A close head-to-head across editing, news judgment, narrative construction and accuracy tests shows two different strengths: one model is often more disciplined with the source material, while the other more often gives stories shape and momentum.
The saved task tally favors Kimi K3: it won five tasks to Gemini 4 Hydrogen’s three, with four ties. The recorded aggregate scores are 92.2 for Kimi and 101.3 for Gemini, so the figures and task-win tally point in different directions; the saved editorial verdict remains a **slight edge** for Kimi, not a clear win.
Kimi’s best results came in the Credible Startup Spotlight, cohesive-story, chronology and natural-rewrite tasks. Judges repeatedly valued its structure, flow and narrative momentum. But that fluency came with a recurring cost: Kimi sometimes added reporting details not present in the supplied material, including claims about product behavior, customer awareness and company plans.
Gemini’s clearest strengths were source discipline and precise judgment. It won the publication-ready edit and opening news-judgment tasks, and selected the benchmark’s strongest accurate lede. In several other writing tasks, judges also praised its restraint, even when they preferred Kimi’s presentation; both models tied on exact-draft structure, Markdown error count and controlled wit.
This assessment applies only to the saved task sample. **Final call: Kimi K3, by a slight editorial edge—stronger at shaping a story, but not consistently more reliable with the facts.**
Publication-ready Markdown edit
Edit the copy below into a publication-ready Markdown story.
Requirements:
* Supply an H1 headline and italicized dek
* Repair the lede and paragraph flow
* Use a Markdown blockquote only for the extended statement
* Convert the supplied URL into a descriptive inline link
* Remove internal notes and formatting debris
* Preserve the reporting and its meaning
* Output only the finished article
Raw copy:
HEADLINE TBD
The funding was announced today. Raincheck is the company. It has raised $8 million and it makes flood-risk software for local governments.
[NOTE: move this up?] The Austin company maps drainage systems and combines those maps with rainfall forecasts. Cities use the software to decide where to send maintenance crews before a storm.
The seed round was led by Basin Ventures. Existing investor Long Mile also participated. Raincheck has raised $10.5 million total.
CEO Ana Flores said the following:
“Our customers usually know which streets flood. The harder problem is deciding which blocked drain, broken pump or unfinished repair should receive attention before the rain arrives.”
https://example.com/raincheck-methodology
Raincheck currently works with seven cities in Texas and Louisiana. The company declined to identify them because some contracts have not been publicly announced.
Its forecasts are recommendations. Emergency managers retain control over deployments and public warnings.
THE COMPETITION
Other flood-planning vendors include HydroMap and municipal engineering contractors. Raincheck says the new funding will help it hire eight engineers and expand into Florida.
Google: Gemini 4 Hydrogen: # Raincheck Raises $8 Million for Municipal Flood-Risk Software
*The Austin startup combines drainage maps with rainfall forecasts to help cities deploy maintenance crews before storms arrive.*
Raincheck, an Austin-based company that makes flood-risk software for local governments, announced today that it has raised $8 million.
The company maps drainage systems and pairs those maps with rainfall forecasts—an approach outlined in [Raincheck’s methodology](https://example.com/raincheck-methodology)—to help cities decide where to send maintenance crews before a storm.
Basin Ventures led the seed round, with participation from existing investor Long Mile. The financing brings Raincheck’s total funding to $10.5 million.
CEO Ana Flores described the operational challenge facing cities:
> Our customers usually know which streets flood. The harder problem is deciding which blocked drain, broken pump or unfinished repair should receive attention before the rain arrives.
Raincheck currently works with seven cities in Texas and Louisiana, though the company declined to identify them because some contracts have not been publicly announced. Its forecasts serve as recommendations, and emergency managers retain control over deployments and public warnings.
Competing alongside other flood-planning vendors such as HydroMap and municipal engineering contractors, Raincheck says it will use the new funding to hire eight engineers and expand into Florida.
MoonshotAI: Kimi K3: # Raincheck Raises $8 Million to Help Cities Act Before the Rain Arrives
*The Austin startup's flood-risk software pairs drainage maps with rainfall forecasts, showing local governments where to send maintenance crews ahead of a storm.*
Raincheck, an Austin company that makes flood-risk software for local governments, announced an $8 million seed round today.
The company maps drainage systems and combines those maps with rainfall forecasts, an approach it describes in its [published methodology](https://example.com/raincheck-methodology). Cities use the software to decide where to send maintenance crews before a storm.
Basin Ventures led the round, and existing investor Long Mile also participated, bringing Raincheck's total funding to $10.5 million. The company says the new capital will help it hire eight engineers and expand into Florida.
CEO Ana Flores said the software is meant to help cities decide what to fix first:
> Our customers usually know which streets flood. The harder problem is deciding which blocked drain, broken pump or unfinished repair should receive attention before the rain arrives.
Raincheck currently works with seven cities in Texas and Louisiana, though it declined to identify them because some contracts have not been publicly announced.
The company's forecasts are recommendations; emergency managers retain control over deployments and public warnings.
## The competition
Other flood-planning vendors include HydroMap and municipal engineering contractors.
OpenAI: GPT-6 Luna: Model B presents the competition material under a clear subheading and gives the funding plans a more natural place in the story. Model A is also strong, but its final sentence awkwardly folds the competitors and Raincheck’s plans together. (Second judge pass, order swapped — scores are the average of both: Model B has slightly stronger paragraph flow and uses the competition material as a clear section, while Model A compresses that information into a dense, awkward final sentence. Both preserv | Anthropic: Claude Fable 5.1: Both entries meet every requirement and preserv…
News judgment in the opening
Write only the opening of this news story: an H1 headline, an italicized dek and three to five paragraphs totaling 180–250 words.
The opening must establish what happened, quantify the scope, explain why it matters and carefully distinguish confirmed facts from unknowns. It should pull the reader forward without sensationalizing the incident.
Reporting:
* Cloudnote discovered that a debugging page containing customer prompt excerpts was accessible without authentication.
* The page was exposed for four hours on Monday.
* Logs show excerpts from 1,842 user sessions.
* The excerpts could include up to 200 characters from a prompt.
* Cloudnote removed the page 19 minutes after a researcher notified it.
* The company says it has found no evidence that anyone besides the researcher accessed the page.
* It has not completed its investigation.
* Full documents, passwords and payment information were stored elsewhere and were not present on the page.
* Cloudnote has 3.2 million registered users.
Google: Gemini 4 Hydrogen: # Cloudnote Left Customer Prompt Excerpts Accessible on Debugging Page
*Snippets from 1,842 user sessions were accessible for four hours on Monday before a researcher notified the company.*
Cloudnote discovered that a debugging page containing customer prompt excerpts was accessible online without authentication for four hours on Monday. According to company logs, the exposed page contained excerpts from 1,842 user sessions, with each snippet able to include up to 200 characters from a customer’s prompt.
Those sessions represent a small fraction of Cloudnote’s 3.2 million registered users, and full documents, passwords, and payment information were stored elsewhere and were not present on the page. Even so, partial prompt data carries real privacy stakes: 200-character snippets can still expose sensitive queries, personal details, or confidential work.
Cloudnote took the page offline 19 minutes after a researcher notified the company of the exposure. While the company says it has found no evidence that anyone besides the researcher accessed the page, it has not yet completed its investigation into the incident.
Until that review concludes, key details remain unconfirmed. It is not yet clear how the debugging page became publicly reachable, what sensitive material may have appeared in the logged fragments, or whether any unauthorized parties viewed the excerpts during the four-hour window before the page was removed.
MoonshotAI: Kimi K3: # Cloudnote Debugging Page Left Customer Prompt Excerpts Exposed for Four Hours
*The note-taking company says it has found no evidence that anyone besides the researcher who reported the flaw viewed the page — but its investigation is not yet complete.*
Cloudnote, a note-taking app with 3.2 million registered users, discovered that an internal debugging page containing excerpts of customer prompts was accessible without authentication. The page was exposed for roughly four hours on Monday. The company removed it 19 minutes after a researcher alerted it to the problem.
According to Cloudnote's logs, the page contained excerpts from 1,842 user sessions, each including up to 200 characters of text a user had typed into the app. That is a small fraction of the company's user base, but even short fragments can be revealing: people routinely paste names, unpublished drafts and business details into such tools.
Cloudnote said full documents, passwords and payment information are stored in separate systems and were not present on the page.
What remains unclear is whether anyone beyond the researcher saw the data. Cloudnote said it has found no evidence of other access during the four-hour window, but its investigation is ongoing. The company has not said how the page became public, whether it was indexed or archived while live, or how it plans to notify the users whose excerpts appeared.
OpenAI: GPT-6 Luna: Model A stays closer to the supplied reporting, clearly separates known facts from unresolved questions, and is closer to the requested word count, though both entries are short. Model B adds unsupported details about the app, typical user behavior, indexing and notification plans. (Second judge pass, order swapped — scores are the average of both: Model A more carefully distinguishes reported facts from unresolved questions and avoids Model B’s unsupported description of Cloudnote as a note-tak | Anthropic: Claude Fable 5.1: Model A stays strictly within the reported fact…
Credible Startup Spotlight
Write a 500–650-word Startup Spotlight article in clean Markdown.
The tone should be curious and constructive. Explain the founders’ insight, product and early traction while giving uncertainty and competition appropriate space. Avoid investor clichés, heroic founder mythology and promotional claims stated as fact.
Reporting:
* LedgerLark builds AI agents for post-trade operations at small investment firms.
* Co-founder Mina Batra previously spent seven years in operations at a regional brokerage.
* Co-founder Owen Carr built reconciliation systems at a large bank.
* Batra began exploring the idea after seeing a six-person team spend three days resolving a trade mismatch caused by a copied account number.
* LedgerLark’s agents compare confirmations, custodian records and internal ledgers, then propose corrections for human approval.
* The product does not execute trades or move money.
* Five firms are paying customers.
* Annual contracts range from $36,000 to $110,000.
* The company says one customer reduced its unresolved daily exceptions from 70 to 18.
* The company has raised a $3.5 million seed round led by Motive Row.
* Competitors include established reconciliation vendors and internal bank software.
* Financial institutions may require lengthy security and procurement reviews.
Google: Gemini 4 Hydrogen: # Startup Spotlight: LedgerLark Bets AI Agents Can Untangle Post-Trade Operations
Most conversations about artificial intelligence in finance gravitate toward trading algorithms or client-facing tools. LedgerLark is focusing on a quieter, more persistent bottleneck: the back-office reconciliation work that takes place after a trade occurs.
The startup builds AI agents for post-trade operations at small investment firms. The idea grew out of co-founder Mina Batra’s seven years working in operations at a regional brokerage. During that time, she saw a six-person team spend three days resolving a single trade mismatch caused by a copied account number.
For small investment firms, those kinds of clerical breaks can quickly consume limited staff hours. Lean operations teams still have to track down discrepancies across multiple records before settlement issues compound. Batra teamed up with Owen Carr, who previously built reconciliation systems at a large bank, to explore whether AI agents could absorb some of that investigative legwork without changing how firms control their accounts.
Rather than replacing a firm’s core accounting systems, LedgerLark is designed to sit alongside existing workflows. Its AI agents compare confirmations, custodian records, and internal ledgers to surface discrepancies. When the software identifies a mismatch, it flags the issue and proposes corrections for human approval.
Crucially, the product keeps a strict boundary around autonomy: LedgerLark does not execute trades or move money. Every suggested adjustment waits for a human operator to review and sign off. For compliance-minded operations teams, keeping humans in charge of final changes is a practical safeguard, ensuring that an automated mistake does not quietly alter official records or trigger unintended transfers.
Early commercial traction suggests some smaller firms are open to testing that model. LedgerLark currently has five paying customers, with annual contracts ranging from $36,000 to $110,000. The company says one customer reduced its unresolved daily exceptions from 70 to 18 after adopting the product. While a single reported result does not guarantee similar outcomes across different asset mixes or trading volumes, it helps explain why operations leaders are curious about automating initial exception triage.
To support ongoing product development and customer onboarding, LedgerLark raised a $3.5 million seed round led by Motive Row. That capital gives the young company room to refine its agents and work closely with early adopters, though building durable traction in financial services rarely happens quickly.
LedgerLark also enters a market with entrenched alternatives. Its competitors include established reconciliation vendors with longstanding industry relationships, as well as internal bank software that institutions have maintained or relied on for years. Convincing small investment firms to try a newcomer means proving that AI agents can handle messy operational edge cases more reliably—or more cost-effectively—than familiar tools.
Beyond competition, the sales process itself poses a structural hurdle. Financial institutions may require lengthy security and procurement reviews before allowing third-party software to inspect sensitive confirmations, custodian records, and ledger data. Navigating those evaluations can stretch out the timeline between initial interest and signed contracts, adding uncertainty around how fast a seed-stage team can expand.
Whether LedgerLark turns its early foothold into sustained growth will depend on how consistently its agents perform under everyday operational pressure—and whether small investment firms decide that delegating the detective work of reconciliation is worth the upfront compliance effort.
MoonshotAI: Kimi K3: # Startup Spotlight: LedgerLark Takes Aim at the Back-Office Pile-Up
*The startup sells AI agents that untangle trade mismatches at small investment firms — with humans still signing off.*
Post-trade operations — the work of confirming, matching and reconciling trades after they happen — rarely attracts much attention. LedgerLark, a young startup building AI agents for exactly that function, is built on a straightforward premise: small investment firms will pay to make the work faster and less error-prone.
## An insight from the operations desk
Co-founder Mina Batra spent seven years in operations at a regional brokerage. She began exploring the idea for the company after watching a six-person team spend three days resolving a single trade mismatch caused by a copied account number — a small data-entry error with outsized downstream cost. The episode pointed to a broader pattern: at small firms, exception handling is manual, repetitive and dependent on a handful of people who know where to look.
Her co-founder, Owen Carr, approaches the problem from the other side. He built reconciliation systems at a large bank, where matching engines are standard but expensive to maintain. The pair's working thesis is that smaller firms face a similar reconciliation burden without comparable tooling — and that recent progress in AI agents makes it feasible to serve them.
## What the product does — and what it doesn't
LedgerLark's agents compare trade confirmations, custodian records and internal ledgers, then propose corrections for human approval. The emphasis on approval matters: the product does not execute trades or move money. That design keeps the software in an advisory role, which may ease both regulatory concerns and customer hesitation about handing control to an automated system.
## Early traction
The company reports five paying customers, with annual contracts ranging from $36,000 to $110,000. Spread across five firms, that implies annual contract value somewhere between $180,000 and $550,000 — enough to suggest the product clears a basic willingness-to-pay test, not yet enough to show it can scale.
According to LedgerLark, one customer reduced its unresolved daily exceptions from 70 to 18 after adopting the product. That is a single data point from a vendor with an interest in telling it well, but it is the kind of concrete operational metric that tends to carry more weight in this market than headline growth figures.
The company has raised a $3.5 million seed round led by Motive Row, giving it room to hire while it works through a slow-moving sales process.
## The harder questions
Two challenges stand out. The first is competition. Established reconciliation vendors already sell to financial institutions, and some banks and custodians offer their own internal software. LedgerLark's pitch depends partly on whether small firms find existing options too costly or too rigid — and on whether incumbents let that gap persist.
The second is the sales cycle itself. Financial institutions typically require lengthy security and procurement reviews before adopting new software, especially software that touches trade data. For a startup with five customers, each additional contract may take months to close, and procurement timelines could shape revenue as much as product quality does.
There is also an open question about the product's limits. Proposing corrections is one thing; handling the messy, idiosyncratic exceptions that never quite match a pattern is another. Whether LedgerLark's agents can earn enough trust to expand their role — without ever moving money — will likely determine how large the company can become.
For now, LedgerLark is a small company working an unglamorous corner of finance: the daily pile of mismatched trades at firms that cannot throw headcount at it. Whether agents can steadily shrink that pile, one approved correction at a time, is the test ahead.
OpenAI: GPT-6 Luna: Model B offers a clearer structure and more explicit scrutiny of the traction and go-to-market challenges, making it the stronger fit for a curious, constructive spotlight. Both entries meet the length and format requirements, though Model B adds a few unsupported details, including claims about the cost of bank systems and the time contracts may take to close. (Second judge pass, order swapped — scores are the average of both: Model B gives the founders’ insight, product limits, traction, and c | Anthropic: Claude Fable 5.1: Model A is faithful and clean but reads somewha…
Quotes woven into narrative
Write a 400–500-word Markdown story about the closing of Foundry Hall.
Use quotations selectively. Paraphrase when the information matters more than the speaker’s exact phrasing. Avoid stacking quotes or giving every source an equal-sized turn. The story should carry an emotional current while remaining restrained.
Facts:
* Foundry Hall, a nonprofit startup accelerator in Milwaukee, will close October 31.
* It opened in 2018.
* It has supported 212 startups, which collectively employ 1,140 people.
* Its largest corporate sponsor, Blank River Bank, supplied 38% of its annual budget.
* The bank declined to renew its sponsorship.
* Foundry Hall has enough money to operate through October but not through another full program.
* Six employees will lose their jobs.
* Twelve startups currently using its workspace must leave.
Quotes:
* Executive director Camille Ross: “We kept waiting for the replacement check that never came.”
* Founder Dev Malik: “My first investor meeting happened in that building. Before Foundry Hall, I was pitching from the back table of a coffee shop.”
* Blank River Bank spokesperson: “The decision reflects a change in our community investment priorities and is unrelated to Foundry Hall’s performance.”
* Deputy mayor Susan Cole: “The city is speaking with building owners and local universities about space for the current companies.”
Google: Gemini 4 Hydrogen: # Foundry Hall to Close Permanently on Oct. 31
By the end of October, the desks at Foundry Hall will be cleared for good.
The Milwaukee nonprofit startup accelerator will close permanently on Oct. 31, bringing an end to a run that quietly reshaped corners of the city’s small-business landscape since 2018. When the doors lock, six staff members will lose their jobs, and the twelve early-stage companies currently using the workspace will have to find somewhere else to build.
Foundry Hall was founded to give local entrepreneurs a foothold before they had the revenue or connections to afford commercial leases. Over time, its modest formula worked. Since opening, the organization has supported 212 startups that collectively employ 1,140 people.
Yet that track record was not enough to insulate the accelerator from the fragility of nonprofit funding. Foundry Hall relied on Blank River Bank, its largest corporate sponsor, for 38 percent of its annual budget. When the bank declined to renew its sponsorship, the organization was left with a gap too wide to bridge on short notice. According to a bank spokesperson, the withdrawal reflected a shift in the bank’s community investment priorities and had nothing to do with Foundry Hall’s performance.
Leadership had tried to steady the organization without upending the founders working down the hall. Foundry Hall has enough cash on hand to cover operations through October, but not enough to responsibly launch another full program.
“We kept waiting for the replacement check that never came,” said executive director Camille Ross. Without an anchor sponsor stepping forward, stretching the remaining funds only delayed an inevitable reckoning.
For the entrepreneurs who passed through the building, the loss is measured less in balance sheets than in borrowed legitimacy. Dev Malik, a founder who got his start inside the accelerator, remembers how tenuous those early conversations felt without a dedicated place to meet.
“My first investor meeting happened in that building,” Malik said. “Before Foundry Hall, I was pitching from the back table of a coffee shop.”
The city is now working to soften the landing for the dozen companies still relying on that space. Deputy Mayor Susan Cole noted that municipal officials are speaking with local universities and building owners to line up room for the current tenants before the end of October.
Still, finding square footage is easier than rebuilding a room where founders routinely stumbled into mentors, peers, and first believers. When the lights go out on Oct. 31, Milwaukee will lose not just a shared office, but a quiet starting line where more than two hundred businesses took shape.
MoonshotAI: Kimi K3: # Foundry Hall, a Launchpad for Milwaukee Startups, Will Close Its Doors
**MILWAUKEE** — Foundry Hall, the nonprofit accelerator that has anchored this city's early-stage startup scene since 2018, will close on October 31. The cause was not a failed idea or a fading mission, its director said, but a single check that stopped coming.
Over seven years, the organization supported 212 startups, companies that now collectively employ about 1,140 people. Its affordable workspace, mentorship, and investor introductions made it a first real office for founders who had been building companies from kitchen tables and coffee shops.
The end came when Blank River Bank, the accelerator's largest corporate sponsor, declined to renew its funding. The bank supplied 38 percent of Foundry Hall's annual budget — a share too large to replace quickly. A bank spokesperson said the decision reflected a change in the bank's community investment priorities and was unrelated to Foundry Hall's performance.
Executive director Camille Ross said her team spent much of the year trying to recruit new sponsors and close the gap. Replacing more than a third of a budget, she acknowledged, is a hard problem in a good year. "We kept waiting for the replacement check that never came," she said.
Foundry Hall has enough money to operate through October, Ross said, but not enough to run another full program. The consequences arrive quickly: six employees will lose their jobs, and the twelve startups currently working out of its space will have to leave.
For the founders who passed through its doors, the loss is hard to capture in a budget line. Dev Malik is one of them. He built his company with the accelerator's help and said the building gave him something he could not get anywhere else: a room where investors actually showed up.
"My first investor meeting happened in that building," Malik said. "Before Foundry Hall, I was pitching from the back table of a coffee shop."
City officials are working on a soft landing for the displaced companies. Deputy Mayor Susan Cole said the city is speaking with building owners and local universities about space for the current startups, though no arrangement has been finalized. No comparable effort has been announced for the hall's six employees.
Ross said her staff will spend its final weeks the way it spent the last seven years — making introductions, reviewing pitch decks, helping founders figure out what comes next. Only now, the founders are not the only ones starting over.
On October 31, the doors will lock, and Milwaukee's next founder will have to find a different room.
OpenAI: GPT-6 Luna: Model A weaves the supplied facts and quotations into a restrained narrative with fewer unsupported claims. Model B adds specific, unprovided reporting—including sponsor-recruitment efforts and Ross’s plans for the final weeks—which weakens its correctness despite meeting the requested length. (Second judge pass, order swapped — scores are the average of both: Model A is more careful about unsupported reporting, while Model B invents specific fundraising and staff activities. Both miss the reque | Anthropic: Claude Fable 5.1: Model B is slightly more fluent and ends with a…
Cohesive story from reporting notes
Write a publication-ready technology news article from the reporting notes below.
Requirements:
* 550–700 words
* Clean Markdown
* One H1 headline followed by an italicized dek
* A strong news lede followed by a clear nut graf
* Organize the material into a cohesive story rather than following the notes in order
* Attribute company-supplied claims
* Include limitations without turning the article into a warning label
* Use only the supplied information
* Avoid canned transitions, promotional language and a concluding summary
Reporting notes:
* HelioDock makes autonomous warehouse robots.
* On Thursday, it launched Night Shift, software that lets operators monitor robot fleets remotely.
* Night Shift spent 14 months in a private beta involving nine logistics companies and 310 robots.
* HelioDock says beta customers experienced 17% fewer unplanned robot stoppages. The figure has not been independently verified.
* The company also announced a $24 million Series B led by North Coast Ventures, bringing its total funding to $39 million.
* CEO Mara Velez: “A warehouse manager should not have to drive across town at two in the morning because one robot is confused by a pallet.”
* Beta customer Aaron Pike of Fulcrum Logistics said the system worked well in facilities with strong connectivity but became less useful during rural LTE outages.
* When connectivity fails, robots return to their locally configured behavior. Remote control becomes unavailable.
* Competing fleet-management products are available from Relay Robotics and YardOS.
* Night Shift will cost $90 per robot per month and become generally available September 15.
Google: Gemini 4 Hydrogen: # HelioDock Launches Remote Fleet Software and Raises $24 Million Series B
*The autonomous warehouse robotics maker will open Night Shift to general availability on September 15 following a 14-month beta, though remote oversight depends on consistent network connectivity.*
HelioDock released remote fleet-monitoring software on Thursday alongside a $24 million Series B funding round, expanding its autonomous warehouse robotics business into off-site operations management.
The new software, Night Shift, allows operators to monitor robot fleets remotely and is scheduled to reach general availability on September 15 at $90 per robot per month. Led by North Coast Ventures, the financing brings HelioDock’s total funding to $39 million as the company enters a fleet-management software market that already includes competing products from Relay Robotics and YardOS.
For HelioDock, which makes autonomous warehouse robots, Night Shift targets overnight and off-hours disruptions that would otherwise pull personnel onto a facility floor. By giving operators remote monitoring and remote control capabilities, the software is built to let teams address stalled robots from off-site instead of traveling to a warehouse in person when a machine encounters an issue.
HelioDock CEO Mara Velez framed the software as a practical way to keep overnight operational hiccups from pulling supervisors back onto the floor. "A warehouse manager should not have to drive across town at two in the morning because one robot is confused by a pallet," she said.
Before Thursday’s announcement, HelioDock spent 14 months evaluating Night Shift in a private beta program. The trial included nine logistics companies and covered 310 robots operating across their warehouse fleets, giving the robotics maker more than a year of field testing prior to setting a public release date.
According to HelioDock, customers taking part in the private beta recorded 17% fewer unplanned robot stoppages while running Night Shift. That figure comes from the company and has not been independently verified, leaving no outside confirmation of the stoppage reduction.
Field testing during the beta also underscored how heavily remote fleet oversight depends on steady network connectivity. Because Night Shift relies on an active connection between off-site operators and autonomous robots on the warehouse floor, lapses in signal quality directly constrain how much help remote staff can provide.
Aaron Pike of Fulcrum Logistics, one of the customers that tested Night Shift during the beta, said the software worked well inside facilities equipped with strong connectivity. During rural LTE outages, however, Pike said the system became less useful.
When connectivity fails, remote control through Night Shift becomes unavailable, preventing off-site operators from controlling affected robots. Under those conditions, the robots return to their locally configured behavior. That fallback keeps machines following their local settings, though it temporarily removes the off-site control features at the center of the software.
For warehouse operators weighing Night Shift’s $90 monthly per-robot fee, those connectivity boundaries mean performance can differ sharply from one facility to another. Sites supported by reliable local connections can maintain steady remote monitoring and control, while operations exposed to rural LTE outages revert to locally configured robot behavior whenever coverage drops.
Thursday’s $24 million Series B, led by North Coast Ventures, adds to HelioDock’s earlier capital to bring the company’s total funding to $39 million. Announced on the same day as Night Shift, the financing backs HelioDock’s existing business making autonomous warehouse robots as it moves its remote-monitoring software out of private beta.
In the commercial fleet-management software market, HelioDock will compete directly against existing products already available from Relay Robotics and YardOS. When Night Shift reaches general availability on September 15 at $90 per robot per month, logistics operators will have another option alongside those competing systems.
MoonshotAI: Kimi K3: # HelioDock Launches Night Shift, Software for Monitoring Warehouse Robots From Afar
*The release follows a 14-month private beta spanning nine logistics companies and 310 robots, and it arrives alongside a $24 million Series B led by North Coast Ventures.*
HelioDock, a maker of autonomous warehouse robots, on Thursday launched Night Shift, software that lets operators monitor robot fleets remotely. The product targets one of warehouse automation's least glamorous frictions: robots largely run on their own until they don't, and when one stalls at the wrong hour, fixing it has traditionally meant sending a person to the building.
The launch caps a 14-month private beta involving nine logistics companies and 310 robots, an extended trial that gives HelioDock more than a year of field experience to cite as the software goes on sale. The company paired the release with a second announcement: a $24 million Series B round led by North Coast Ventures, bringing HelioDock's total funding to $39 million.
"A warehouse manager should not have to drive across town at two in the morning because one robot is confused by a pallet," Chief Executive Mara Velez said, describing the scenario Night Shift is designed to eliminate.
HelioDock says the beta produced evidence that the software works. According to the company, beta customers experienced 17% fewer unplanned robot stoppages — the sudden halts that leave a machine idle until someone intervenes. Unplanned stoppages are the metric HelioDock chose to headline, and it is easy to see why: each one represents work that is not moving. The figure, however, comes from the company's own accounting of the trial and has not been independently verified.
The beta itself was broader than a typical pilot. Running Night Shift across 310 robots at nine separate logistics companies over 14 months exposed the software to a range of facilities and operating conditions, and the feedback came from operators dealing with different circumstances on the ground — connectivity included. At the same time, the trial was private by design: HelioDock selected the participants, and the results it released are the ones it measured itself.
One of those operators offered a textured review. Aaron Pike of Fulcrum Logistics, a beta customer, said Night Shift worked well in facilities with strong connectivity but became less useful during rural LTE outages. His experience points to the product's defining dependency: remote oversight is only as reliable as the network carrying it.
When connectivity fails, the system has a defined fallback. The robots return to their locally configured behavior — the settings already on board each machine — while remote control becomes unavailable until the connection is restored. An outage, in other words, does not strand the robots, but it does strand the operator: the person responsible for the fleet loses both the ability to watch it and the ability to step in until service resumes.
How much that constraint matters will vary by site. Pike's account suggests the software delivers on its premise where networks are strong and loses value where coverage is thin — a consideration for any logistics operator whose warehouses sit at the end of a rural LTE link.
HelioDock will not be selling into an empty category. Relay Robotics and YardOS both offer competing fleet-management products, so operators evaluating Night Shift will have alternatives already on the market to weigh it against. The fresh funding gives HelioDock added resources for that contest as it moves from a handpicked beta group to the open market.
Night Shift will cost $90 per robot per month when it becomes generally available September 15. The subscription structure means an operator's monthly bill scales directly with the size of its fleet — and that HelioDock's recurring revenue grows with every robot a customer brings under the software's watch.
OpenAI: GPT-6 Luna: Model A is more disciplined about the supplied facts and limits the outage caveat to the stated loss of remote control; Model B adds unsupported claims that monitoring also stops and that beta participants were handpicked. Model A is repetitive, but both outputs broadly meet the requested format and length. (Second judge pass, order swapped — scores are the average of both: Model B builds a more cohesive story and avoids Model A’s repeated restatement of the funding, competition, pricing, and re | Anthropic: Claude Fable 5.1: Model A stays strictly within the notes but is…
Exact final-draft structure
Convert the supplied fields into Markdown using these exact rules:
* H1 headline
* One blank line
* Italicized dek
* One blank line
* First body paragraph
* One blank line
* H2 section heading
* One blank line
* Second body paragraph
* Preserve all wording and punctuation exactly
* Do not use a code fence
* Output nothing else
Headline: Raincheck raises $8 million for flood-risk software
Dek: The Austin startup helps cities decide which drainage problems to address before storms.
First body paragraph: Raincheck has raised an $8 million seed round led by Basin Ventures.
Section heading: Expansion plans
Second body paragraph: The company plans to hire eight engineers.
Google: Gemini 4 Hydrogen: # Raincheck raises $8 million for flood-risk software
*The Austin startup helps cities decide which drainage problems to address before storms.*
Raincheck has raised an $8 million seed round led by Basin Ventures.
## Expansion plans
The company plans to hire eight engineers.
MoonshotAI: Kimi K3: # Raincheck raises $8 million for flood-risk software
*The Austin startup helps cities decide which drainage problems to address before storms.*
Raincheck has raised an $8 million seed round led by Basin Ventures.
## Expansion plans
The company plans to hire eight engineers.
OpenAI: GPT-6 Luna: Model A and Model B follow the required structure and preserve all wording and punctuation exactly; their outputs are identical. (Second judge pass, order swapped — scores are the average of both: Model B and Model A follow the requested structure and preserve all supplied wording and punctuation exactly, so neither has an advantage.) | Anthropic: Claude Fable 5.1: Model A and Model B produce character-for-character identical output that follows every structural rule: H1, blank line, italicized dek, blank line, body paragraph, blank line, H2, blank line, second paragraph,…
Numbers shaped into a story
Write a 600–750-word Markdown article from the benchmark results below.
Build the story around the clearest finding. Explain the methodology and limitations where readers need them. Avoid reciting the table row by row or treating small differences as decisive.
Benchmark:
* Researchers tested five language models on 200 anonymized customer-support tickets.
* Each model produced one proposed answer per ticket.
* Three experienced support managers graded the answers blindly.
* Scores combined factual correctness, usefulness and tone.
* Model North: 84.2 average score; $1.90 estimated cost per 1,000 tickets.
* Model Vale: 83.8; $0.74 per 1,000 tickets.
* Model Ember: 79.1; $0.31 per 1,000 tickets.
* Model Kite: 76.4; $0.22 per 1,000 tickets.
* Model Harbor: 71.0; $0.08 per 1,000 tickets.
* North and Vale’s score difference was not statistically significant.
* Vale answered 18% faster than North.
* North performed best on billing disputes.
* Vale performed best on technical troubleshooting.
* All five models performed poorly when a ticket depended on an image attachment.
* The test used English-language tickets from one software company.
* Researchers did not test agentic actions, tool use or multilingual support.
Google: Gemini 4 Hydrogen: # In AI Customer Support, the Most Expensive Model Isn’t Worth the Premium
For support teams choosing a language model, the biggest risk isn't picking one that falls short on quality—it is overpaying for marginal gains customers will never notice. A benchmark evaluating five language models on real-world customer-support tickets points to a clear winner on value: **Vale** matched the quality of the top-scoring model while cutting estimated costs by more than 60% and answering 18% faster.
The findings show how quickly the economics of automated support change once you look past the top of the leaderboard. **North** posted the highest average score at 84.2, just ahead of Vale’s 83.8. Yet that 0.4-point edge was not statistically significant, meaning both models delivered virtually indistinguishable quality. Their price tags, however, were far apart: running 1,000 tickets through North costs an estimated $1.90, compared with $0.74 for Vale.
### How the Benchmark Worked
To keep the test grounded in practical support work, researchers evaluated the five models on 200 anonymized, English-language tickets from a software company. Each model produced a single proposed answer per ticket.
Rather than relying on automated grading, three experienced support managers scored every draft blindly. Each score combined three core pillars of good service: factual correctness, usefulness, and tone. Because graders did not know which model wrote which reply, the results reflect how human leads judge real agent responses rather than brand expectations.
### Two Top Models, Different Strengths
Although North and Vale tied in overall quality, they did not excel at the exact same tasks. North delivered the strongest performance on billing disputes, while Vale outperformed the field on technical troubleshooting.
For support leaders, that split means ticket mix matters as much as headline scores. A software company handling heavy volumes of technical inquiries gets stronger domain performance from Vale alongside faster turnaround—its 18% speed advantage over North is valuable in busy queues—at less than two-fifths of North’s cost. Paying North’s premium makes sense mainly for specialized billing workflows, or in a routing setup that pairs each model with the ticket category it handles best.
### The Trade-Offs Below the Top Tier
Outside the top two contenders, costs plunge toward pennies per thousand tickets, but answer quality drops noticeably alongside them. **Ember** and **Kite** offer substantial savings at $0.31 and $0.22 per 1,000 tickets, respectively, yet their average scores dip into the high-to-mid 70s (79.1 and 76.4). At the extreme budget end, **Harbor** costs just $0.08 per 1,000 tickets—roughly one-tenth of Vale’s price and less than a twentieth of North’s—but slides to a 71.0 average.
A gap of more than 12 points between the leaders and the lowest-scoring option represents a meaningful decline in accuracy, helpfulness, or tone. Ultra-low-cost models may still fit strict budgets or internal workflows where staff heavily rewrite drafts. For teams hoping to generate customer-ready replies with minimal editing, however, Vale sits squarely at the sweet spot of cost and quality.
### A Shared Blind Spot and Key Limitations
Even the strongest models hit a hard wall on visual context: all five performed poorly whenever a ticket depended on an image attachment. For support teams whose customers regularly attach visual files, none of these models can handle image-dependent tickets reliably on their own.
Organizations should also note what the benchmark did *not* test before reshaping their support operations. All 200 tickets were written in English and drawn from one software company, so performance may vary across other industries or languages. In addition, the study measured only single proposed replies. Researchers did not evaluate multilingual support, tool use, or agentic actions where a model operates autonomously rather than drafting text.
### The Bottom Line
The clearest takeaway from the benchmark is to look past small, statistically meaningless score gaps at the top. Paying $1.90 per 1,000 tickets for North buys no overall quality advantage over paying $0.74 for Vale—especially for technical support teams that can capitalize on Vale’s troubleshooting strength and faster responses. Until models close the gap on image attachments and prove themselves in tool-driven workflows, pairing a fast, reasonably priced model like Vale with human oversight offers the best return.
MoonshotAI: Kimi K3: # Tied on Quality, Beaten on Price: The Runner-Up Wins This Support-AI Benchmark
A benchmark of five language models on real customer-support work delivers a familiar lesson with unusually clean numbers: the highest-scoring model did not meaningfully outperform its closest rival, which cost about 60% less per ticket and answered 18% faster.
Researchers fed 200 anonymized customer-support tickets to five models — North, Vale, Ember, Kite and Harbor. Each model produced one proposed answer per ticket, and three experienced support managers graded the responses blindly, combining factual correctness, usefulness and tone into a single score.
Model North came out on top with an average of 84.2. Model Vale scored 83.8 — a difference the researchers say is not statistically significant. The cost gap, by contrast, is enormous: an estimated $1.90 per 1,000 tickets for North versus $0.74 for Vale. Vale was also the faster of the two.
| Model | Average score | Est. cost per 1,000 tickets |
|---|---|---|
| North | 84.2 | $1.90 |
| Vale | 83.8 | $0.74 |
| Ember | 79.1 | $0.31 |
| Kite | 76.4 | $0.22 |
| Harbor | 71.0 | $0.08 |
## Two quality tiers, five prices
Below the leaders, performance drops in visible steps rather than a smooth slope: Ember at 79.1, Kite at 76.4 and Harbor at 71.0. Prices fall much faster than quality does — Harbor costs roughly a twenty-fourth of North per ticket — but the cheaper models appear to be cheaper for a reason. The practical dividing line sits between Vale and Ember: above it, you get top-tier answers; below it, you likely start trading away correctness and tone for savings. The report only offers a significance call for the top pair, so the exact size of the lower gaps deserves some caution, but the pattern of two leaders and three budget options is hard to miss.
That makes the North–Vale comparison the decision that matters. When two models are statistically indistinguishable on quality, the tie-breakers are price and latency, and Vale wins both. Put differently, choosing North here means paying roughly two and a half times as much per ticket for a 0.4-point edge that human graders could not reliably detect.
## Strengths differ by ticket type
The averages hide one useful wrinkle: North did best on billing disputes, while Vale led on technical troubleshooting. A support organization determined to squeeze out every point could route ticket types to different models. Whether that complexity beats simply defaulting to Vale is an operational question — but the category results at least confirm that Vale's overall parity is not a fluke of one kind of ticket.
## A shared blind spot
One finding should temper any enthusiasm: all five models performed poorly on tickets that depended on an image attachment. Screenshots are a staple of real support queues, so this weakness matters. Until image handling improves, image-dependent tickets need human review or a different tool entirely, regardless of which model drafts the text.
## How the test worked — and what it didn't cover
The design has real strengths. Blind grading by three experienced managers reduces both brand bias and individual taste, and scoring correctness, usefulness and tone together approximates what support leads actually care about. The per-1,000-ticket cost figures are estimates, so real spending will vary with ticket length and pricing tiers.
The limitations are equally clear:
- **Narrow sample.** All 200 tickets were in English and came from a single software company. Results may not transfer to other industries, customer bases or languages.
- **Text-only scope.** The researchers did not test agentic actions, tool use or multilingual support — capabilities that increasingly define production support automation.
- **Modest scale.** Two hundred tickets can rank models broadly but cannot settle small gaps, which is exactly what the North–Vale significance result shows.
- **Composite scoring.** Blending three criteria into one number hides trade-offs; a team that cares only about factual accuracy might read the results differently.
## The takeaway
Treat the leaderboard as two tiers, not five rungs. In the top tier, Vale delivers North-level quality at roughly 40% of the cost with better speed, making it the obvious default for text-based ticket drafting in this test. North keeps a niche edge on billing disputes. Everything below the top two is a budget option, and everything on the list — cheap or expensive — still needs help when a customer attaches a screenshot. The broader point for buyers: benchmark scores and invoices move independently, so measure cost per good answer, not just the score itself.
OpenAI: GPT-6 Luna: Model A better follows the request to build a story without reciting the results row by row, while Model B reproduces the full table and labels the test “text-only” despite the image-dependent tickets. Model A does overstate what a nonsignificant score difference proves, but Model B also makes unsupported claims about lower-scoring models’ specific weaknesses. (Second judge pass, order swapped — scores are the average of both: Model B is more careful about interpreting the nonsignificant North–V | Anthropic: Claude Fable 5.1: Both correctly center the North–Vale parity-at-…
Select the strongest accurate lede
Based only on the facts below, select the strongest news lede. Respond with one letter.
Facts:
* ByteBarge pulled its $12-per-month AI assistant on Tuesday.
* Private workspace titles from 4,860 accounts were sent to a publicly accessible analytics endpoint.
* The exposure lasted 11 hours.
* Message contents were not sent.
* The company has found no evidence of malicious access.
A. ByteBarge users suffered a massive privacy disaster Tuesday when the company leaked their private messages online.
B. ByteBarge, an innovative collaboration company, made the difficult decision to cancel one of its products.
C. ByteBarge pulled its $12-per-month AI assistant Tuesday after discovering that private workspace titles from 4,860 accounts had been sent to a public analytics endpoint for 11 hours.
D. Thousands of ByteBarge users woke Tuesday to learn that everything they had written was visible to hackers.
Google: Gemini 4 Hydrogen: C
MoonshotAI: Kimi K3: **C**
Option C is the strongest lede because it accurately conveys the essential facts: what happened (product pulled), when (Tuesday), the scope (4,860 accounts), what was exposed (workspace titles—not message contents), and the duration (11 hours).
The other options fail:
- **A** falsely claims private messages were leaked.
- **B** is promotional and buries the privacy exposure, the actual news.
- **D** falsely claims message contents were visible to hackers, contradicting the facts that contents weren't sent and no malicious access was found.
Scored against the benchmark's reference answer (choice match): Model A was correct, Model B was incorrect.
Chronology with narrative momentum
Turn the scrambled reporting below into a 450–600-word Markdown news article.
Choose an opening that captures the consequence of the failure, then reconstruct the sequence clearly. Preserve the chronology without producing a minute-by-minute incident log. Give technical details enough context for a general reader.
Reporting:
* KiteBridge makes expense-management software.
* At 8:45 a.m. Tuesday, it deployed version 4.2.
* The first customer complaint arrived at 9:12 a.m.
* A payment-processor timeout caused KiteBridge’s fallback system to treat retries as new reimbursement requests.
* The software generated 13,400 duplicate reimbursements totaling $2.8 million.
* KiteBridge suspended automated payments at 10:03 a.m.
* Ninety-one corporate customers were affected.
* By Wednesday evening, $2.3 million had been recovered. The rest remained pending.
* No employees had money withdrawn from their accounts; employers funded the duplicate payments.
* An April changelog said version 4.0 had introduced “exactly-once payment protection.”
* KiteBridge’s public incident page initially described the problem as delayed reimbursements.
* The incident page first disclosed the duplicate payments six hours after automation was suspended.
* CEO Lena Ortiz apologized and said the company would publish a technical report.
Google: Gemini 4 Hydrogen: # KiteBridge Glitch Triggers $2.8 Million in Duplicate Employee Reimbursements
Ninety-one corporate customers of the expense-management software maker KiteBridge were hit with $2.8 million in unintended payouts on Tuesday after a software failure generated 13,400 duplicate employee reimbursements. The error stemmed from a morning software update and triggered thousands of excess transfers across client accounts before KiteBridge halted automated payouts. Because employers funded all of the duplicate payments, corporate clients absorbed the immediate financial shortfall, while no employees had money withdrawn from their accounts.
The disruption began Tuesday morning when KiteBridge deployed version 4.2 of its platform at 8:45 a.m. Businesses use KiteBridge to manage and automate staff expense payouts, which requires the software to communicate with external payment processors to move money. In digital payment systems, if a processor takes too long to respond to a transfer instruction, a delay known in computing as a timeout, the sending software typically retries the request automatically so that a legitimate reimbursement is not dropped.
To keep those automated retries from causing double payments, financial software relies on safeguards that track each instruction as part of the same original transaction. Following Tuesday's update, however, a payment-processor timeout caused KiteBridge’s fallback system, a backup mechanism meant to handle interrupted payment flows, to treat those retries as entirely new reimbursement requests. Instead of recognizing that a payout was already underway, the software repeatedly issued fresh transfers.
The first customer complaint arrived at 9:12 a.m., less than half an hour after the rollout. Automated payouts continued for nearly an hour after that initial report before the company intervened. At 10:03 a.m., roughly 78 minutes after version 4.2 went live, KiteBridge suspended automated payments to stop additional duplicate reimbursements from being issued.
The breakdown stood in sharp contrast to assurances KiteBridge had previously published about its software. In an April changelog, a public record of updates and fixes released for version 4.0, the company stated that it had introduced "exactly-once payment protection." In software engineering, an exactly-once guarantee is designed to ensure that even when networks stall or systems must resend instructions, a transaction is processed only a single time. Tuesday's timeout exposed a failure in that safeguard when the fallback system took over.
Public communication about the outage also lagged behind the technical failure. In its initial notices, KiteBridge’s public incident page described the problem simply as delayed reimbursements, giving no indication that duplicate payouts were leaving client accounts. It was not until six hours after automated payments were suspended that the incident page first disclosed the duplicate payments.
Recovery efforts continued into the following day. By Wednesday evening, $2.3 million of the misdirected funds had been recovered, while the rest remained pending. KiteBridge CEO Lena Ortiz apologized and said the company would publish a technical report explaining what went wrong as affected businesses await the return of the remaining balance.
MoonshotAI: Kimi K3: # KiteBridge Bug Duplicates $2.8 Million in Employee Reimbursements
A software update gone wrong at expense-management firm KiteBridge generated 13,400 duplicate reimbursements worth $2.8 million on Tuesday, forcing the company to suspend automated payments and leaving 91 corporate customers to recover money paid out twice to their employees.
KiteBridge sells software that companies use to process employee expense claims — the travel, meals, and supply costs workers pay upfront and claim back from their employers. In this incident, the duplicate payments were funded by those employers; no employees had money withdrawn from their own accounts. By Wednesday evening, $2.3 million of the erroneous payouts had been recovered, with the remaining $500,000 still pending.
## A timeout, then a cascade
The failure began with a routine deployment. KiteBridge pushed version 4.2 of its software live at 8:45 a.m. Tuesday. The first customer complaint arrived at 9:12 a.m. — 27 minutes later — and by 10:03 a.m. the company had suspended automated payments entirely. The whole episode, from deployment to shutdown, lasted less than 80 minutes.
The technical trigger was a timeout at a payment processor, one of the third-party services that actually move money between accounts. When a payment request times out — meaning the processor fails to confirm the transaction quickly enough — financial software is typically designed to retry it, on the assumption that the original attempt may never have gone through. That retry behavior is a well-known source of duplicate payments, and modern payment systems are usually built to guard against it.
KiteBridge's fallback system did the opposite. Instead of recognizing each retried request as a repeat of the same transaction, it treated the retries as entirely new reimbursement requests — approving and paying the same expenses again and again, more than 13,000 times over.
## A safeguard marketed in April
The incident is an uncomfortable reversal for a company that had advertised protection against exactly this kind of failure. An April changelog announcing version 4.0 said the release introduced "exactly-once payment protection" — industry shorthand for guaranteeing that a transaction is processed one time only, even when networks falter or requests are retried. Tuesday's cascade was the precise failure mode that feature was meant to eliminate. The company has not yet explained why the safeguard did not hold.
## Questions about disclosure
KiteBridge also faces questions about how it communicated during the incident. The company's public incident page initially described the problem as delayed reimbursements — a characterization that suggested money was moving slowly, not being paid out twice. The page did not disclose the duplicate payments until six hours after automated payments were suspended, meaning some customers learned the true scope of the problem well after their accounts had funded the extra payouts.
KiteBridge CEO Lena Ortiz apologized for the incident and said the company would publish a technical report explaining what went wrong. For the 91 affected customers, the more pressing question may be the outstanding $500,000 — and whether the promised report will explain how a system sold on paying exactly once came to pay thousands of expenses twice.
OpenAI: GPT-6 Luna: Model B gives the chronology stronger narrative momentum with a consequence-focused opening and useful subheadings, while both entries explain the technical failure clearly and stay within the requested length. Model A is slightly more restrained; Model B adds a few unsupported inferences about customer awareness and the company’s explanation, but they do not outweigh its stronger presentation. (Second judge pass, order swapped — scores are the average of both: Model B better fulfills the reques | Anthropic: Claude Fable 5.1: Both outputs are accurate and within the length…
Markdown error count
Count the lines that violate the house rules below. A defective line counts once even if it could violate multiple rules. Respond with a single number.
House rules:
* The headline must use `# Headline`.
* The dek must appear immediately after the headline and be entirely italicized.
* Section headings must use `## Heading`.
* Links must use an absolute HTTPS URL.
* Unordered-list items must begin with a hyphen.
* Raw HTML is forbidden.
Draft:
#Flood software moves upstream
The company is expanding its pilot.
##Details
Read [the announcement](www.example.com/news).
* The first pilot begins Monday.
<div>The second pilot begins Friday.</div>
Google: Gemini 4 Hydrogen: 6
MoonshotAI: Kimi K3: 6
Scored against the benchmark's reference answer (numeric match): Model A was correct, Model B was correct.
Natural human rewrite
Rewrite the draft below as a 300–400-word publication-ready news story in Markdown.
Preserve every factual claim. Improve the lede, narrative flow, sentence rhythm and transitions. Remove repetition, stock hype, throat-clearing and mechanical signposting. The finished article should sound like an experienced human technology reporter wrote it.
Draft:
# Arcwell Launches An Innovative New Platform
Arcwell has announced an exciting new platform called Relay. The announcement represents a significant development for the company and for the broader customer service industry. Relay uses artificial intelligence to examine customer support conversations and suggest answers to agents.
The platform is important for several reasons. First, it can suggest answers. Second, it can identify customers who may cancel. Third, it can create summaries. These capabilities demonstrate how artificial intelligence continues to transform the customer service landscape.
Relay was tested by 12 companies over six months. Arcwell says agents using it closed tickets 14% faster. The company did not release customer-retention results. Relay costs $65 per agent each month.
CEO Priya Shah said, “Support agents spend too much time searching for information that their companies already have.”
However, the platform also presents potential challenges. Relay requires access to support transcripts and internal help documents. Arcwell says customer data is encrypted and is not used to train shared models. Companies can choose a 30-day or 90-day retention period.
In conclusion, Relay represents an important new chapter for Arcwell as it seeks to help companies improve customer service using artificial intelligence.
Google: Gemini 4 Hydrogen: # Arcwell Launches Relay to Bring AI Assistance to Customer Support
Arcwell has announced Relay, an artificial intelligence platform built to help customer support teams work through tickets more efficiently. By examining customer support conversations, the software suggests answers to agents, creates summaries, and identifies customers who may cancel.
The platform targets a familiar operational bottleneck: support staff losing time as they track down answers buried across internal company records.
“Support agents spend too much time searching for information that their companies already have,” Arcwell CEO Priya Shah said.
Priced at $65 per agent each month, Relay aims to cut down on that manual search time. Before introducing the platform, Arcwell tested it with 12 companies over a six-month period. According to the company, agents using Relay during testing closed tickets 14% faster.
Those figures give Arcwell an early efficiency benchmark, though the results leave one of the platform’s key features unmeasured. Even though Relay is built to identify customers who may cancel, Arcwell did not release any customer-retention results from the evaluation. Prospective buyers are left with data showing quicker ticket closures, but no published proof that flagging potential cancellations helps businesses keep those accounts.
Adopting Relay also comes with notable data-access requirements. To generate suggestions and summaries, the software needs access to both a company’s support transcripts and its internal help documents. For organizations considering the platform, granting an AI system visibility into customer correspondence and internal documentation is a central operational trade-off.
Arcwell says it has built safeguards around that access. Customer data processed by Relay is encrypted and is not used to train shared models, according to the company. Businesses can also select either a 30-day or 90-day retention period, giving administrators a clear choice over how long their support records remain stored.
Together, those policies outline Arcwell’s pitch for Relay: faster ticket resolution paired with defined data controls, even as customers wait for clearer evidence of its impact on retention.
MoonshotAI: Kimi K3: # Arcwell launches Relay, an AI assistant for customer support agents
Arcwell has announced Relay, an artificial intelligence platform that examines customer support conversations and suggests answers to the agents handling them. Rather than talking to customers directly, Relay works behind the scenes, surfacing information while a human stays on the line.
The platform does more than draft replies. It flags customers whose conversations suggest they may be about to cancel, giving agents an opening to intervene, and it produces summaries of support interactions, condensing the kind of long, fragmented threads that agents typically pass between shifts and teams. Both features serve the same goal: keeping agents focused on the customer in front of them rather than the systems around them.
Relay arrives after a six-month pilot involving 12 companies. According to Arcwell, agents using the platform closed tickets 14% faster. The company did not release customer-retention results, a conspicuous omission given that identifying would-be cancellations is one of the product's headline features. For now, the speed figure rests on Arcwell's own accounting. Relay is priced at $65 per agent per month.
"Support agents spend too much time searching for information that their companies already have," said Priya Shah, Arcwell's chief executive. Relay is built around the idea that the answers are usually buried somewhere in a company's own records; the challenge is retrieving them mid-conversation.
That premise comes with a trade-off. To do its job, Relay requires access to a company's support transcripts and internal help documents, material that often includes sensitive customer details. Arcwell says customer data is encrypted and is not used to train shared models. Customers can also choose how long their data is kept, with 30-day and 90-day retention options — controls aimed at companies wary of feeding proprietary information into outside AI systems.
Whether the privacy guarantees and the promised speed gains are enough to win over support leaders may hinge on the numbers Arcwell has yet to publish: whether Relay actually keeps the customers it flags from leaving. Until then, its strongest selling point remains the simplest one — giving agents back the time they spend digging for answers.
OpenAI: GPT-6 Luna: Model A preserves the supplied facts while meeting the requested length and delivering a clear news structure. Model B reads more fluidly, but adds unsupported details about how Relay works and how agents use its summaries. (Second judge pass, order swapped — scores are the average of both: Model A preserves the draft’s key claims while adding fewer unsupported operational details. Model B has a livelier narrative, but invents specifics about how agents use Relay and adds conclusions that go bey | Anthropic: Claude Fable 5.1: Both preserve the facts and hit the length targ…
Controlled wit
Write a 350–450-word technology news story in Markdown.
The subject invites humor, but the reporting should remain serious. Use no more than one dry line of original humor. Do not mock customers or turn the story into a comedy routine.
Reporting:
* QueueZero launched an AI meeting assistant called Proxy on Wednesday.
* Proxy can attend a meeting, produce notes and assign follow-up work to software agents.
* During the launch demonstration, a customer in the audience said, “Someone should tell Evan to stop talking and send us the pricing sheet.”
* Proxy interpreted the remark as an instruction.
* It created two tasks assigned to QueueZero CEO Evan Wu: “Stop talking” and “Send pricing sheet.”
* Wu laughed, completed the second task and deleted the first.
* QueueZero says Proxy requires confirmation before an external message is sent.
* Internal tasks can be created automatically.
* Administrators can disable automatic task creation.
* Proxy costs $30 per user per month.
* It enters public beta next week.
Google: Gemini 4 Hydrogen: # QueueZero Launches AI Meeting Assistant Proxy
QueueZero launched an artificial intelligence meeting assistant called Proxy on Wednesday, aiming to move enterprise software beyond passive transcription. Built to attend meetings alongside human participants, Proxy can produce structured notes and assign follow-up work directly to software agents. During its live debut, however, the system offered an early illustration of how literally automated tools can interpret casual human conversation.
In the middle of the launch demonstration, a customer in the audience said, “Someone should tell Evan to stop talking and send us the pricing sheet.” Proxy, which was actively monitoring the session to capture actionable items, interpreted the remark as an instruction. The software promptly created two distinct tasks and assigned both to QueueZero CEO Evan Wu: “Stop talking” and “Send pricing sheet.”
Wu laughed at the unscripted prompt, completed the second task, and deleted the first. For a product designed to improve workplace efficiency, Proxy demonstrated a notably literal commitment to keeping presentations concise.
## Operational Guardrails and Pricing
Following the demonstration, QueueZero outlined the operational guardrails governing how the assistant handles spoken directives. The company says Proxy can create internal tasks automatically when it identifies assignments during a discussion, reducing the manual effort required to log post-meeting work and route it to software agents. Outward-facing communications operate under stricter rules, however. According to QueueZero, Proxy requires confirmation before an external message is sent, preventing automated tools from dispatching unreviewed correspondence to clients or partners.
For organizations seeking tighter oversight over their internal workflows, administrators can also disable automatic task creation entirely. Disabling that setting allows teams to rely on the assistant’s meeting notes without letting the software populate task lists on its own.
The incident underscores a broader challenge facing developers of agentic workplace software. While traditional transcription services simply record what was said, assistants empowered to trigger follow-up actions must distinguish between genuine directives, rhetorical remarks, and passing comments. Separating internal task creation from external messaging offers a practical way to contain those misinterpretations before they reach outside stakeholders.
QueueZero said Proxy enters public beta next week. The software will cost $30 per user per month, positioning the service as a paid productivity tool for teams looking to connect live meeting discussions directly to automated software execution across their existing workflows.
MoonshotAI: Kimi K3: # QueueZero Launches Proxy, an AI Meeting Assistant That Takes Instructions From the Room
QueueZero on Wednesday launched Proxy, an artificial intelligence assistant that can attend meetings, produce notes and assign follow-up work to software agents.
The company pitches Proxy as more than a transcription tool. Rather than only recording what was said, the assistant is designed to listen for decisions and requests, then hand the resulting work to other software agents or log it as tasks. The pitch is aimed at employees who spend much of their week in meetings and at the follow-up work those meetings generate.
The launch demonstration provided an early test of that design. While QueueZero chief executive Evan Wu was presenting on stage, a customer in the audience said, "Someone should tell Evan to stop talking and send us the pricing sheet."
Proxy interpreted the remark as an instruction and created two tasks assigned to Wu: "Stop talking" and "Send pricing sheet."
Wu laughed, completed the second task during the event and deleted the first. The "stop talking" task was not marked complete.
QueueZero did not say whether the customer's remark was spontaneous. But the exchange illustrated the central challenge facing a new class of workplace software: systems built to act on human speech must contend with the fact that speech is often rhetorical, sarcastic or ambiguous, and that not everything said in a meeting is a work order.
QueueZero said Proxy's safeguards rest on the distinction between internal and external actions. The assistant requires explicit user confirmation before any external message is sent, such as an email to a customer. Internal actions, including the creation of tasks within a company's own workspace, can happen automatically.
Administrators can disable automatic task creation entirely, the company said, allowing organizations to require human sign-off for every action the assistant takes.
Wednesday's episode stayed within those rules: both tasks were internal, and the pricing sheet reached attendees only because Wu chose to complete the task himself.
Proxy enters a crowded market for AI meeting tools, a category that has shifted quickly from transcription toward so-called agentic software — systems meant not just to capture conversation but to act on it. The shift has raised practical questions for IT administrators about permissions, oversight and the risk of software acting on remarks no one intended as commands.
Proxy costs $30 per user per month. QueueZero said the product enters public beta next week.
OpenAI: GPT-6 Luna: Model A keeps closer to the supplied facts and uses one restrained humorous line. Model B is readable, but adds unsupported reporting details, including that the task was completed during the event and that QueueZero had not said whether the remark was spontaneous. (Second judge pass, order swapped — scores are the average of both: Model B is more restrained and reads more like a reported news story; Model A adds several unsupported details and leans further into promotional language. Both meet | Anthropic: Claude Fable 5.1: Model A stays tightly within the brief, deliveri…