Head to head: Qwen: Qwen3.8 Max (0902) vs Anthropic: Claude Fable 5.1

Qwen: Qwen3.8 Max (0902): # HelioDock Launches Night Shift Remote Monitoring for Warehouse Robots and Raises $24 Million *The software emerges from a 14-month private beta as HelioDock says fleets saw fewer unplanned stoppages and announces new funding 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, and announced a $24 million Series B led by North Coast Ventures. The announcement pairs a product debut with new capital for a company selling automation management to logistics operators. Night Shift is meant to let operators watch robots from a distance and reduce the need for a manager to respond in person when a robot stops unexpectedly. HelioDock says beta customers experienced 17% fewer unplanned robot stoppages. The figure has not been independently verified. Night Shift spent 14 months in a private beta involving nine logistics companies and 310 robots. That period gave the company a group of outside fleets to test the software before the general release. The funding announced Thursday brings HelioDock’s total to $39 million. CEO Mara Velez described the product as a way to avoid late-night site visits for minor robot issues. “A warehouse manager should not have to drive across town at two in the morning because one robot is confused by a pallet,” Velez said. Night Shift’s remote monitoring role places operational visibility away from the warehouse floor. Operators can watch fleet activity from another location, while the robots retain local instructions for use when a connection is absent. The software’s value is tied to the connection between the operator and the fleet; when that link disappears, the remote functions disappear with it. The company is selling Night Shift in a market where competing fleet-management products are available from Relay Robotics and YardOS. HelioDock’s pricing is $90 per robot per month, with general availability set for September 15. 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. That limitation is central to a product designed to keep operators connected to robots from another location. When connectivity fails, robots return to their locally configured behavior. Remote control becomes unavailable. The comments from Fulcrum Logistics show how network quality can shape the usefulness of remote fleet supervision. A warehouse with dependable service may allow an operator to monitor many robots without being on the floor. A rural site with intermittent LTE may lose the remote layer that Night Shift provides, even if the robots continue operating under their local settings. The beta underscores the operational split created by remote supervision. In facilities with strong connectivity, Night Shift can keep a manager informed about robot status without a physical inspection. During an LTE outage, the same facility may have to rely on local robot behavior if intervention is needed. HelioDock’s beta numbers are company-supplied and part of its launch argument, but the company does not provide independent verification for the 17% reduction in unplanned stoppages. The 17% figure covers nine logistics companies and 310 robots during the 14-month beta test, not a broader set of deployments. The figure does not show whether the result was uniform across all beta customers or varied by facility, robot count, or network condition. North Coast Ventures led the $24 million Series B, which brings HelioDock’s total funding to $39 million. Night Shift will become generally available on September 15 at $90 per robot per month. At that rate, a ten-robot deployment would cost $900 per month.

Anthropic: Claude Fable 5.1: # HelioDock Launches Remote Fleet Monitoring Software, Raises $24 Million *Night Shift lets warehouse operators oversee autonomous robots from afar, though a beta customer says its usefulness depends on the connection.* HelioDock, a maker of autonomous warehouse robots, on Thursday introduced Night Shift, software that lets operators monitor and manage robot fleets remotely, and announced a $24 million Series B funding round to support the launch. The product is aimed at a recurring problem in automated warehouses: a single robot stalls or misreads its surroundings, and someone has to be physically present to sort it out. Night Shift is meant to let a manager diagnose and address those situations from a laptop rather than from the warehouse floor. It will be generally available September 15 at a price of $90 per robot per month. "A warehouse manager should not have to drive across town at two in the morning because one robot is confused by a pallet," HelioDock CEO Mara Velez said. ## Fourteen months in the field Night Shift arrives after a 14-month private beta involving nine logistics companies and 310 robots. HelioDock says customers in that program saw 17 percent fewer unplanned robot stoppages. The company did not provide the underlying data, and the figure has not been independently verified. Aaron Pike of Fulcrum Logistics, one of the beta customers, offered a mixed assessment. The system worked well in facilities with strong connectivity, he said, but became less useful during rural LTE outages. That limitation follows from how the software is designed. Night Shift depends on a live link between the operator and the fleet. When that link fails, the robots fall back to whatever behavior has been configured locally on each machine, and remote control is unavailable until the connection is restored. In a facility with reliable network coverage, that fallback may rarely come into play. In a distribution center at the edge of cellular range, it could happen at exactly the moment a manager most wants to intervene. HelioDock has not said whether it plans to address connectivity gaps in future versions of the product. ## A crowded category Remote fleet management is not a new idea in warehouse robotics. Relay Robotics and YardOS both sell competing products, and HelioDock will be entering a market where operators already have options. The company's pitch rests partly on its own hardware: Night Shift is built to work with HelioDock's robots, and the 310 units in the beta program gave the company a sizable installed base on which to test it before the public launch. The per-robot pricing model means costs scale directly with fleet size. A facility running 50 HelioDock robots would pay $4,500 a month for the software; one running 300 would pay $27,000. ## New funding The Series B was led by North Coast Ventures and brings HelioDock's total funding to $39 million. The company did not disclose a valuation or say how the proceeds would be allocated beyond supporting the Night Shift rollout. The round roughly doubles the capital HelioDock has raised to date, and the timing suggests investors are backing the company's move from selling hardware alone toward recurring software revenue. At $90 per robot per month, a customer with the same 310 robots that participated in the beta would generate about $335,000 a year in subscription fees, a meaningful figure for a company of HelioDock's size. Velez did not say how many of the nine beta customers have committed to paying for Night Shift once the free trial period ends. Fulcrum Logistics, for its part, has not said publicly whether it will continue. For warehouse operators weighing the product, the question Pike raised may be the most practical one: not whether remote monitoring is useful in principle, but whether the network in a given building is good enough to make it useful in practice. HelioDock's own description of the fallback behavior makes clear that in the places where connectivity is weakest, Night Shift hands control back to the robots themselves.

Qwen: Qwen3.8 Max (0902): # Startup Spotlight: LedgerLark Tames Post-Trade Paperwork for Small Investment Firms LedgerLark is building artificial intelligence agents for a part of finance that rarely attracts attention: the back-office work that begins after a trade is executed. For small investment firms, that work often means comparing confirmations, custodian records and internal ledgers to find where numbers no longer match. The startup’s founders came to the problem from operations roles. Co-founder Mina Batra spent seven years in operations at a regional brokerage. She began exploring the idea after seeing a six-person team spend three days resolving a trade mismatch caused by a copied account number. Co-founder Owen Carr built reconciliation systems at a large bank, where similar issues were often handled through a combination of custom software, spreadsheets and manual checks. LedgerLark’s product reflects that experience. Its agents review post-trade documents and records, identify discrepancies and propose corrections for human approval. The company emphasizes that the software does not execute trades or move money. That boundary is important in a sector where firms are unlikely to hand autonomous systems control over client assets, especially early in adoption. The approach aims to reduce what operations teams often call “exceptions”: items that fail to match and require investigation. Instead of asking staff to scan every record, the agents rank issues and draft possible fixes. A human then reviews the recommendation and decides whether to accept it. This keeps the system in a support role, which may make it easier for compliance-conscious firms to test. Early traction is modest but specific. LedgerLark says it has five paying customers, with annual contracts ranging from $36,000 to $110,000. The company also reports that one customer reduced unresolved daily exceptions from 70 to 18. As with most early-stage metrics, that figure should be treated cautiously: results may depend on the customer’s existing processes, data quality and the complexity of its trades. Still, for a small operations team, a reduction of that size could mean fewer late-night checks and less month-end pressure. The company has raised a $3.5 million seed round led by Motive Row. The funding should give LedgerLark time to refine its agents, strengthen integrations and add controls expected by financial institutions. It is not yet clear how quickly the startup can expand beyond its initial customer set. Sales in financial services often move slowly, and firms may require lengthy security reviews, procurement processes and vendor approvals before allowing new software to touch operational data. Competition is another open question. Established reconciliation vendors already serve parts of this market, and many larger banks rely on internal software built over years. LedgerLark’s focus on smaller investment firms may help it avoid direct confrontations with the largest providers, but those firms still have alternatives, including spreadsheets, consultants and existing platforms adding AI features. The startup will need to show that its agents are not merely a smarter dashboard, but a tool that consistently reduces manual work without creating new risks. That may be the central test for LedgerLark. Post-trade operations are not glamorous, but errors can be expensive and time-consuming. If the company can prove reliability across different custodians, asset classes and reporting formats, it may earn a durable place in small firms’ daily workflows. If not, its agents could become another pilot that stalls when security, integration or accuracy concerns arise. For now, LedgerLark looks like a careful attempt to apply AI where it can be useful rather than dramatic. The founders are not promising to replace traders or overhaul markets. They are trying to make a tedious, high-stakes checking process faster and more manageable. In financial operations, that may be enough.

Anthropic: Claude Fable 5.1: # Startup Spotlight: LedgerLark Wants to Take the Drudgery Out of Trade Reconciliation Every trade a small investment firm makes leaves behind a paper trail that has to agree with itself. The confirmation from the counterparty, the record held by the custodian and the firm's own internal ledger should all say the same thing. Often they do. When they don't, someone has to find out why. LedgerLark, a young company building AI agents for post-trade operations, is betting that this work is both more painful and more automatable than it looks from the outside. ## The problem, seen up close Co-founder Mina Batra spent seven years in operations at a regional brokerage. The moment she points to as the origin of the company was unglamorous: a six-person team spent three days resolving a single trade mismatch. The cause turned out to be an account number that had been copied incorrectly. Three days of skilled labour for a typo is the kind of inefficiency that persists not because anyone thinks it's acceptable, but because the alternative—rebuilding the tooling—feels out of reach for a firm that size. Large banks have dedicated reconciliation infrastructure. Smaller shops often rely on spreadsheets, vendor exports and institutional memory. Batra's co-founder, Owen Carr, comes from the other end of that spectrum. He built reconciliation systems at a large bank, which gives the pair a useful contrast: one knows what the problem looks like without resources, the other knows what the expensive solutions actually do. ## What the product does—and doesn't do LedgerLark's agents ingest confirmations, custodian records and internal ledgers, compare them, and flag discrepancies. Where the system can infer a likely cause and fix, it proposes a correction. A human then approves or rejects it. The boundary the company has drawn matters. The product does not execute trades and does not move money. It is positioned as a recommendation layer rather than an autonomous one, which is probably the only sensible place to start in a domain where errors are costly and regulators are attentive. Whether customers eventually want more autonomy, and whether LedgerLark would be comfortable offering it, remains an open question. ## Early traction Five firms are currently paying customers, with annual contracts ranging from $36,000 to $110,000. LedgerLark has raised a $3.5 million seed round led by Motive Row. The most concrete result the company cites is one customer reducing its unresolved daily exceptions from 70 to 18. That figure comes from LedgerLark rather than independent verification, and a single customer is a small sample. Still, if it holds across deployments, it would represent a meaningful change in how an operations team spends its day. ## Where it gets harder The competitive landscape is not empty. Established reconciliation vendors have been selling into this market for decades, and large institutions frequently build their own software in-house. LedgerLark's wager is that the small-firm segment is underserved by both—too small for enterprise vendors to prioritise, too resource-constrained to build internally. That seems plausible, but incumbents can move downmarket, and "AI agent" features are easy to announce. The sales process is another obstacle. Financial institutions, even small ones, tend to run lengthy security and procurement reviews before letting a third party near their trade data. For a seed-stage company with five customers, each deal may take months, and the cost of those cycles can eat into a modest runway. There is also the question of trust in the agents themselves. A system that proposes corrections is only valuable if its proposals are right often enough that reviewing them is faster than doing the work from scratch. The 70-to-18 figure suggests that threshold is being met in at least one case. Proving it holds across different custodians, asset classes and data quality levels is the work ahead. ## Worth watching for LedgerLark is attacking a real, specific problem with founders who have seen it from both sides. The next signals to look for are customer count beyond the initial five, evidence of renewals, and whether the company can shorten the procurement path that stands between it and a larger market.

Qwen: Qwen3.8 Max (0902): # The Last Lights in Foundry Hall For seven years, Foundry Hall has sounded like a place with a future: keyboards tapping after evening workshops, founders rehearsing pitches in glass-walled rooms, the nervous laughter of people trying to turn an idea into payroll. That sound will end on October 31, when the Milwaukee nonprofit startup accelerator closes its doors. The closure follows the loss of its largest corporate sponsor. Blank River Bank, which supplied 38% of Foundry Hall’s annual budget, chose not to renew its sponsorship. The bank said the decision reflected a shift in community investment priorities and was not tied to the accelerator’s performance. But the funding gap proved too wide to close in time. Foundry Hall has enough money to operate through October, not long enough to carry another full program. Executive director Camille Ross said the staff kept waiting for “the replacement check that never came.” By the time it was clear no rescue funding would arrive, the calendar had already turned against them. Since opening in 2018, Foundry Hall has supported 212 startups, which collectively employ 1,140 people. Those numbers are the kind that appear in grant applications and city reports, but they began as smaller, human moments. Founder Dev Malik remembers his first investor meeting taking place inside the building, a long way from the back table of a coffee shop where he once tried to pitch his company. Now, twelve startups using the workspace must find somewhere else to go. Six employees will lose their jobs. The companies are not shutting down all at once, but they are being asked to leave a place that gave them structure, advice, and a shared address at a vulnerable stage. City officials are trying to soften the landing. Deputy mayor Susan Cole said the city is talking with building owners and local universities about possible space for the current companies. Those conversations may help some startups survive the transition, but they do not replace the organization itself. Foundry Hall was never just square footage. It was proof that Milwaukee could make room for people building something new, often without much margin for error. Its closing leaves a quiet floor, an unfinished program, and a reminder that civic ecosystems depend on money that can disappear in a single renewal cycle. In its final weeks, the hall will keep doing what it was built to do: advising founders, helping teams plan their next move, and trying to send them out with more than a goodbye. The lights will stay on until October 31. After that, the city will have to decide whether the next generation of startups can find another place to begin.

Anthropic: Claude Fable 5.1: # Foundry Hall Will Close Its Doors October 31 For six years, Foundry Hall has been the place where Milwaukee's founders went to stop improvising. The nonprofit startup accelerator, which opened in 2018, announced this week that it will shut down at the end of October. The decision follows the loss of its largest corporate sponsor, Blank River Bank, which supplied 38 percent of the organization's annual budget and declined to renew when its commitment came up this year. Foundry Hall has enough money to keep operating through October. It does not have enough to run another full program, and leadership concluded that stretching a fraction of a budget across a half-formed cohort would serve no one well. Six employees will lose their jobs. Twelve startups currently working out of the building will have to find somewhere else to be. "We kept waiting for the replacement check that never came," said executive director Camille Ross. She spent much of the year approaching other companies and foundations, she said, and found plenty of sympathy but no one willing to step in at the scale the bank had. By late summer, the math was simply not there. In a statement, Blank River Bank said the decision reflected a change in its community investment priorities and was unrelated to Foundry Hall's performance. The bank did not elaborate on what those priorities now are. By the record the accelerator leaves behind, performance was never the question. Since 2018, Foundry Hall has supported 212 startups. Together, those companies now employ 1,140 people. Dev Malik is one of those founders. He still describes the place less in terms of what it gave him than in terms of what it let him stop doing. "My first investor meeting happened in that building," he said. "Before Foundry Hall, I was pitching from the back table of a coffee shop." That is the part of the loss that resists a spreadsheet. Desk space can be rented. A conference room can be borrowed. What Foundry Hall offered was a room where a first-time founder was expected to show up, and where investors expected to find them when they did. City officials say they are trying to soften the landing for the twelve companies still in the building. Deputy mayor Susan Cole said the city is speaking with building owners and local universities about space for them, though she offered no timeline and no assurance that the group could stay together. For now, the building remains open. Founders still badge in each morning. Ross said the staff's remaining job is to get the current companies through the next few months with as little disruption as possible, and to write down what the organization learned so it does not leave with the furniture. Then, on the last day of October, the doors will lock.

Qwen: Qwen3.8 Max (0902): # Cloudnote Exposes Debugging Page With Customer Prompt Excerpts for Four Hours *The company says there is no evidence anyone besides a security researcher viewed the page, but the investigation remains open.* Cloudnote said Monday that a debugging page containing excerpts from customer prompts was publicly accessible without authentication for four hours, exposing a limited slice of user activity on its platform. The company said server logs show the page included excerpts from 1,842 user sessions. Each excerpt contained as many as 200 characters from a prompt, meaning the exposed material could reveal fragments of users’ queries but not full documents. Cloudnote removed the page 19 minutes after an outside security researcher reported it. The company said it has found no evidence that anyone other than the researcher accessed the page, but it has not completed its investigation. It also said passwords, payment information and full documents were stored elsewhere and were not present on the exposed page. The incident affects a fraction of Cloudnote’s 3.2 million registered users, though the company has not said whether affected users will be notified directly. Cloudnote also has not disclosed what types of prompts were involved. The episode underscores how even temporary development or debugging endpoints can leak sensitive user-generated content if left unprotected, and it leaves open whether any data was copied before its discovery.

Anthropic: Claude Fable 5.1: # Cloudnote Exposed Excerpts of Customer Prompts on Unprotected Debugging Page *A debugging page left open for four hours on Monday displayed fragments from 1,842 user sessions. The company says only the researcher who reported it appears to have viewed the page, but its investigation is not finished.* Cloudnote said an internal debugging page containing short excerpts of customer prompts was reachable without authentication for about four hours on Monday, exposing fragments of what users had typed into the service. According to the company, logs show the page displayed excerpts from 1,842 user sessions. Each excerpt could contain up to 200 characters of a prompt — enough, in some cases, to indicate what a person was asking the service to do, but not the full text. Cloudnote has 3.2 million registered users, so the affected sessions represent a small fraction of its base. A security researcher notified Cloudnote of the exposure, and the company said it removed the page 19 minutes later. Full documents, passwords and payment information were stored in separate systems and were not present on the page, Cloudnote said. The company said it has found no evidence that anyone other than the researcher accessed the page while it was exposed. It also said its investigation is not complete, and it has not said how the page became publicly accessible, which users were affected or whether those users will be notified. Even brief prompt fragments can be sensitive. People routinely type names, health details and confidential work matters into online tools, expecting that text to remain private.

Qwen: Qwen3.8 Max (0902): # Raincheck Raises $8 Million to Help Local Governments Prepare for Floods *The Austin company maps drainage systems and pairs them with rainfall forecasts to guide crews before storms strike.* Raincheck, a company that makes flood-risk software for local governments, announced today that it has raised $8 million in seed funding. The Austin company maps drainage systems and combines those maps with rainfall forecasts, giving cities a way to decide where to send maintenance crews before a storm. Raincheck currently works with seven cities in Texas and Louisiana, but declined to identify them because some contracts have not been publicly announced. The seed round was led by Basin Ventures, with participation from existing investor Long Mile. Raincheck has raised $10.5 million to date. Chief Executive Officer Ana Flores said: > “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.” The company describes its approach in [Raincheck’s methodology](https://example.com/raincheck-methodology). Its forecasts are recommendations, while emergency managers retain control over deployments and public warnings. Other flood-planning vendors include HydroMap and municipal engineering contractors. Raincheck said the new funding will help it hire eight engineers and expand into Florida.

Anthropic: Claude Fable 5.1: # Raincheck Raises $8 Million to Help Cities Clear Drains Before the Storm *The Austin startup's flood-risk software tells local governments where to send maintenance crews ahead of heavy rain.* Raincheck, an Austin company that makes flood-risk software for local governments, announced today that it has raised $8 million in seed funding. The round was led by Basin Ventures, with participation from existing investor Long Mile. It brings Raincheck's total funding to $10.5 million. Raincheck maps municipal drainage systems and combines those maps with rainfall forecasts. Cities use the software to decide where to send maintenance crews before a storm arrives. CEO Ana Flores said the challenge for cities is less about knowing where flooding happens than about choosing 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. The company has published [an explanation of its forecasting methodology](https://example.com/raincheck-methodology). Raincheck currently works with seven cities in Texas and Louisiana. It declined to identify them, saying some contracts have not been publicly announced. The software's forecasts are recommendations; emergency managers retain control over crew deployments and public warnings. Raincheck competes with other flood-planning vendors, including HydroMap, as well as municipal engineering contractors. The company said the new funding will go toward hiring eight engineers and expanding into Florida.

Qwen: Qwen3.8 Max (0902): C

Anthropic: Claude Fable 5.1: C

Qwen: Qwen3.8 Max (0902): # 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.

Anthropic: Claude Fable 5.1: # 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.

Qwen: Qwen3.8 Max (0902): 6

Anthropic: Claude Fable 5.1: **6**

Qwen: Qwen3.8 Max (0902): # Arcwell launches Relay, an AI platform that suggests agent replies and flags churn risk Arcwell has launched Relay, an artificial intelligence platform that examines customer support conversations and suggests answers to agents. The release gives the company a new product aimed at customer service operations, where response times, cancellation signals and document access are central to the pitch. Relay’s features are built around three functions. It suggests answers while agents handle customer conversations, identifies customers who may cancel, and creates summaries of support interactions. By doing so, the platform tries to reduce the amount of time agents spend hunting for information and to surface signals that might otherwise be missed in long queues of tickets and chats. Arcwell said the system was tested by 12 companies over six months. During that period, the company said, agents using Relay closed tickets 14% faster. The company did not release customer-retention results, leaving unanswered the question of whether the churn-warning feature translated into fewer cancellations. Relay is priced at $65 per agent each month. CEO Priya Shah said, “Support agents spend too much time searching for information that their companies already have.” To work, Relay requires access to support transcripts and internal help documents. That access is central to its ability to suggest answers and generate summaries, but it also means companies must consider how customer records are handled. Arcwell said customer data is encrypted and is not used to train shared models. Customers can choose either a 30-day or 90-day retention period. Together, those details show Arcwell is trying to position Relay as more than a text-generation tool. It connects support data, agent productivity and retention signals, while giving administrators controls over how long records remain in the system. The company will still need to show whether faster ticket closure becomes a durable selling point. Without retention results, the clearest evidence so far is the ticket-closure figure.

Anthropic: Claude Fable 5.1: # Arcwell's Relay Puts AI Beside Support Agents, With Retention Results Still Unreported Arcwell has unveiled Relay, an AI platform that reads customer support conversations and proposes replies to the agents handling them. The pitch is straightforward. Agents fielding tickets often know the answer exists somewhere in their company's documentation but lose minutes hunting for it. Relay is meant to close that gap by surfacing suggested responses as a conversation unfolds. "Support agents spend too much time searching for information that their companies already have," said CEO Priya Shah. Suggested answers are the headline feature, but not the only one. Relay also flags customers who appear likely to cancel and generates summaries of support interactions, giving teams a condensed record without requiring agents to write one themselves. ## What the pilot showed — and what it didn't Before launch, Arcwell ran Relay with 12 companies over six months. According to the company, agents using the platform closed tickets 14% faster than they otherwise would have. That figure covers speed alone. Arcwell did not release any results on customer retention, which leaves the cancellation-prediction feature without published evidence of its impact. For buyers weighing the tool, that is a notable gap: faster ticket resolution is easy to measure, but whether Relay actually helps keep customers from leaving remains, for now, an open question. Relay is priced at $65 per agent per month. ## Data access and what Arcwell promises in return To do its job, Relay needs to see a company's support transcripts and its internal help documents — material that can include sensitive customer details and proprietary operational knowledge. That requirement will be a sticking point for some organizations. Arcwell's response is a set of handling commitments. Customer data is encrypted, the company says, and it is not used to train models shared across Arcwell's customer base. Companies also choose how long their data is kept, with a choice between a 30-day and a 90-day retention period. Those assurances address the most common objections to feeding support data into a third-party AI system, though prospective customers will likely want to examine the details before signing on. Relay enters a market where plenty of vendors already promise to make support teams faster. Arcwell's 14% figure gives it a concrete claim to point to. Whether the retention tools deliver comparable, measurable results is something the company has yet to demonstrate.

Qwen: Qwen3.8 Max (0902): # KiteBridge Software Failure Creates $2.8 Million in Duplicate Expense Payments A faulty software update at KiteBridge turned an ordinary expense cycle into a costly double-payment problem for 91 corporate customers, sending out 13,400 duplicate reimbursements worth $2.8 million before the company stopped its automated payment system. The episode began Tuesday morning when KiteBridge, which makes expense-management software, deployed version 4.2 at 8:45 a.m. The first customer complaint arrived at 9:12 a.m., as employees or finance teams saw repeated payments for the same expense requests. At the center of the problem was a timeout involving an outside payment processor. In online payment systems, a timeout can occur when a request is sent but the response takes too long or does not arrive. To keep transactions moving, KiteBridge’s software relied on a fallback system that retried payments that appeared stuck. Because of the flaw introduced or exposed by the update, that fallback treated retries as new reimbursement requests rather than second attempts at the same payment. The result was that one legitimate claim could produce more than one transfer. KiteBridge suspended automated payments at 10:03 a.m., roughly an hour and a quarter after the update went out. By then, the error had already produced thousands of duplicate reimbursements across client organizations. The company said the affected payments were funded by employers, not drawn back from employees’ bank accounts. In practical terms, workers received extra deposits, while the companies using KiteBridge bore the immediate financial exposure. The company’s public communications followed a slower timeline than its internal shutdown. KiteBridge’s incident page initially described the issue as delayed reimbursements. It did not disclose that duplicate payments had been made until six hours after automated payments were suspended. That gap is likely to draw scrutiny from customers, particularly finance departments that rely on quick, accurate explanations when money moves incorrectly. KiteBridge has said it recovered most of the funds quickly. By Wednesday evening, the company had recovered $2.3 million, leaving about $500,000 still pending. The remaining amount may require individual reconciliation with employees, contractors, or company accounts, depending on how the payments were routed and whether recipients had already spent them. The failure also raised questions about earlier claims in the company’s product documentation. In April, KiteBridge’s changelog said version 4.0 had introduced “exactly-once payment protection,” a safeguard intended to prevent the same payment from being issued more than once. The fact that version 4.2 could still generate duplicates suggests either that the protection did not cover the payment path involved in the failure or that later changes weakened it. Chief Executive Lena Ortiz apologized for the incident and said KiteBridge would publish a technical report. For customers, the report will need to explain how a system designed to prevent duplicate payments failed, why the public description initially understated the problem, and what controls will prevent a similar breakdown in future updates.

Anthropic: Claude Fable 5.1: # Software Update at KiteBridge Sends $2.8 Million in Duplicate Reimbursements A routine software release at KiteBridge, a maker of expense-management software, triggered 13,400 duplicate reimbursement payments totaling $2.8 million on Tuesday, leaving 91 corporate customers to foot the bill for money their employees were paid twice. By Wednesday evening, the company said it had recovered $2.3 million. Roughly half a million dollars remained pending. ## How the failure unfolded KiteBridge deployed version 4.2 of its platform at 8:45 a.m. Tuesday. Within half an hour, at 9:12 a.m., the first customer complaint arrived. The trigger, according to the company, was a timeout at a payment processor — the third-party service that actually moves money from an employer's account to an employee's. Timeouts are common in payment systems: a request is sent, no confirmation comes back within the expected window, and the software must decide whether the payment went through. KiteBridge's software is designed to handle this by falling back to a retry. But in version 4.2, that fallback system treated each retry as a brand-new reimbursement request rather than a second attempt at the same one. Every time the system tried again, it created another payment. The result was thousands of reimbursements that were processed, and paid, more than once. The company suspended automated payments at 10:03 a.m., about 78 minutes after the release went live. ## A protection that was supposed to exist The failure is notable because KiteBridge had told customers it had solved exactly this problem. An April changelog accompanying version 4.0 said the release introduced "exactly-once payment protection" — a guarantee, in industry terms, that a single reimbursement would never be paid out more than a single time, even when network hiccups or retries occurred. How that protection broke down in version 4.2 is a question the company has not yet answered publicly. ## Who paid No employees had money pulled from their personal accounts. The duplicate payments were funded by their employers, the 91 companies that use KiteBridge to reimburse staff for expenses. Those employers are now the parties owed money, and the recovery effort is aimed at returning funds to them. ## A slow disclosure KiteBridge's communication about the incident drew its own scrutiny. The company's public incident page initially described the problem as *delayed* reimbursements — a framing that suggested employees were waiting for money, not receiving it twice. The page did not disclose the duplicate payments until six hours after automated payments were suspended, meaning customers checking the status page through most of Tuesday afternoon would have seen an inaccurate description of what had gone wrong. By then, the first affected customer had been in contact with the company for nearly seven hours. ## Apology and next steps Chief Executive Lena Ortiz apologized for the incident and said KiteBridge would publish a technical report explaining the failure. The company has not said when that report will be released, whether version 4.2 remains in production, or when automated payments will resume. For the affected employers, the immediate concern is the outstanding balance. For KiteBridge, the longer question is how a safeguard it advertised in April allowed a single processor timeout to multiply into 13,400 duplicate payments in a little over an hour.

Qwen: Qwen3.8 Max (0902): # For Support Tickets, Vale Matches North at a Fraction of the Cost The clearest takeaway from a new benchmark of five language models for customer-support drafting is not that one model swept the field. It is that Vale delivered nearly the same overall quality as top-scoring North while costing much less and responding faster. For teams evaluating AI support assistants, that makes Vale the most compelling value proposition in this test. The benchmark used 200 anonymized customer-support tickets from a single software company. Each model generated one proposed reply per ticket. Three experienced support managers graded the replies blindly, meaning they did not know which model produced which answer. Scores combined factual correctness, usefulness, and tone — the ingredients that matter most in support: Was the answer right? Did it move the customer toward a solution? Did it sound helpful? The overall scores placed North and Vale close together. North averaged 84.2, while Vale averaged 83.8. That gap was not statistically significant. In practical terms, the benchmark does not show that North produces reliably better support answers than Vale. The rest of the field trailed: Ember scored 79.1, Kite 76.4, and Harbor 71.0. Those differences suggest a tiering, with North and Vale ahead of the lower-cost options, but they should not be read as proof that every point translates into a noticeably better customer experience. Cost changes the story. North is estimated at $1.90 per 1,000 tickets, while Vale costs $0.74 per 1,000 tickets. Vale achieved essentially equivalent quality at roughly 61% lower cost. It was also 18% faster than North, which may matter in queues where response time affects satisfaction. For support organizations weighing quality against budget and latency, Vale’s result is hard to ignore. The cheaper models offer a low-cost alternative, but their scores fall enough to deserve caution. Ember, Kite, and Harbor cost $0.31, $0.22, and $0.08 per 1,000 tickets respectively. They may suit simple, high-volume, low-risk workflows, especially if output is reviewed before sending. However, the benchmark does not show that savings would survive extra human review, rework, or damaged customer trust. In this dataset, the top two models separated themselves from the lower-priced group. One nuance is that overall scores can hide category-level strengths. North performed best on billing disputes, while Vale performed best on technical troubleshooting. That suggests routing tickets by intent could help, but because the test included only 200 tickets total, category results should be treated as directional rather than definitive. A handful of tickets can influence a subgroup average, especially when the overall North-Vale comparison is already too close to declare a winner. The benchmark also reveals a shared weakness: all five models performed poorly when a ticket depended on an image attachment. That is an important reminder that current models may struggle when the customer’s problem cannot be understood from text alone. A screenshot or annotated interface image may contain the crucial detail. If support workflows rely heavily on attachments, human review or a specialized vision pipeline remains necessary. Readers should also keep the test’s boundaries in mind. The tickets were English-language tickets from one software company, so the findings may not generalize to other domains or languages. The benchmark did not test agentic actions, tool use, or multilingual support. It measured draft quality on static tickets, not whether a model can safely look up an account, apply a refund, or handle customers in other languages. Those limits do not make the results unhelpful; they define where the evidence is strongest. For English-language software support drafting, this benchmark points to Vale as the best balance of quality, cost, and speed. North remains a strong option, especially where billing disputes dominate, but the data do not justify treating it as decisively superior. The practical conclusion is that Vale is good enough at a much better price — and organizations should pilot it with their own tickets before paying more.

Anthropic: Claude Fable 5.1: # The Highest Score Didn't Win A new benchmark of five language models on customer-support tickets produced a leaderboard, and the model at the top of it is probably not the one most teams should pick. Model North posted the best average score, 84.2 out of 100. Model Vale came in at 83.8. The gap between them was not statistically significant — the researchers could not distinguish the two on quality. But they could distinguish them on everything else. Vale cost $0.74 per 1,000 tickets to North's $1.90, less than 40 percent of the price, and it returned answers 18 percent faster. That is the story of this benchmark: once quality is held equal, the remaining differences are large and they all point the same direction. ## How the test worked The researchers collected 200 anonymized support tickets from a single software company and had each of the five models produce one proposed answer per ticket. Three experienced support managers then graded every answer without knowing which model wrote it. Each score blended factual correctness, usefulness and tone into a single number. That setup has real strengths. Blind grading by people who actually do the job is closer to how a support team would judge these tools than any automated metric. And the combined score reflects the reality that a technically correct answer delivered rudely, or a warm answer that is wrong, both fail the customer. It also has limits worth keeping in view. Two hundred tickets is enough to separate a 71 from an 84 but not enough to separate an 83.8 from an 84.2 — which is exactly what the significance test says. The tickets came from one company, in one language. Different products, different customer bases and different ticket mixes could shift the results. And the test measured only drafted replies: the researchers did not evaluate agentic actions, tool use or multilingual support, all of which matter in production deployments. ## Reading the rest of the field Below the top pair, the scores drop in steps: Ember at 79.1, Kite at 76.4, Harbor at 71.0. Costs fall faster than quality does — Harbor runs at $0.08 per 1,000 tickets, roughly a twenty-fourth of North's price. Whether those cheaper models are "good enough" depends entirely on what a team does with the output. If a human agent reviews every draft before it goes out, a 79 might be a perfectly reasonable starting point and the savings are real. If the model's answers reach customers directly, a 13-point gap between Harbor and North is likely to show up as repeat contacts and frustrated users. The benchmark cannot settle that question; it can only make the trade-off visible. It is also worth resisting the temptation to read precision into every row. The difference between Ember and Kite is 2.7 points on a 200-ticket sample. That is suggestive, not conclusive. ## Where the two leaders differ North and Vale were not identical in character, even if their averages matched. North did best on billing disputes, where the grading presumably rewarded careful handling of account details and a measured tone. Vale did best on technical troubleshooting. A support operation dominated by one type of ticket might reasonably lean toward the model that handles that type best, and a team with a clear split could route tickets accordingly — though routing itself adds complexity the benchmark did not test. ## The weakness every model shared One result cut across the entire leaderboard: all five models performed poorly when a ticket depended on an image attachment. A customer sending a screenshot of an error, a photo of a damaged product or a picture of a confusing invoice got worse answers from every model, including the two leaders. For a lot of support teams, this is the most practically useful finding in the study. It means no amount of model selection fixes the problem. It has to be handled in the workflow — by routing image-dependent tickets to humans, by asking customers to describe what they see in text, or by adding a separate step that extracts information from attachments before the model drafts a reply. ## What to take away Benchmarks invite a single question: which model is best? This one answers a more useful question: what does quality actually cost, and where does it break down? The honest reading is that two models tied for first, one of them is far cheaper and faster, the cheaper tiers are viable depending on how much human review sits behind them, and none of them should be trusted with a screenshot. That is more actionable than a ranking — and it is a reminder that the top of a leaderboard is a starting point for a decision, not the decision itself.

Qwen: Qwen3.8 Max (0902): # QueueZero’s AI Meeting Assistant Assigns Tasks to CEO After Launch Remark QueueZero on Wednesday launched Proxy, an artificial intelligence meeting assistant designed to attend meetings, produce notes and assign follow-up work to software agents. The launch included a live demonstration that briefly turned toward the company’s own chief executive after Proxy interpreted an audience comment as a task instruction. During the demonstration, a customer in the audience said, “Someone should tell Evan to stop talking and send us the pricing sheet.” Proxy processed the remark and created two tasks assigned to QueueZero chief executive Evan Wu: “Stop talking” and “Send pricing sheet.” Wu laughed after the tasks appeared on screen. He completed the second task and deleted the first, according to the company. The moment demonstrated Proxy’s ability to turn spoken comments into actionable items, a feature that is central to the company’s pitch. QueueZero said Proxy requires confirmation before an external message is sent. Internal tasks, however, can be created automatically, allowing the assistant to record follow-ups and route work without waiting for a person to approve each step. The company said administrators can disable automatic task creation, giving organizations more control over how aggressively the software operates in live meetings. The episode highlights a broader challenge facing makers of AI agents: determining when a statement is a casual remark and when it should be treated as a command. Meeting assistants are being sold as tools that can reduce administrative work by capturing decisions, summarizing discussions and assigning next steps. But automatic task creation also raises questions about accuracy, consent and workplace embarrassment. QueueZero is entering a crowded market for AI productivity tools, where vendors are competing to move beyond chatbots toward agents that can carry out work across software systems. Proxy’s pricing is set at $30 per user per month, and the product is scheduled to enter public beta next week. The company’s launch message emphasized that Proxy can attend meetings on behalf of users, generate notes and delegate follow-up work to software agents. The demonstration added a more specific example of that workflow: a spoken request from the audience became a task list within seconds. For QueueZero, the incident may have served as a useful demonstration of speed. It also showed why settings, confirmations and administrative controls are likely to matter as more companies allow AI systems to listen, write and assign.

Anthropic: Claude Fable 5.1: # QueueZero's New AI Meeting Assistant Assigns Its First Task — to the CEO QueueZero on Wednesday launched Proxy, an AI meeting assistant designed to attend meetings on a user's behalf, generate notes and hand off follow-up work to software agents. The launch demonstration also offered an unplanned look at how literally the system can read a room. Proxy is built to sit in on video calls, transcribe the discussion and identify commitments made during the conversation. Rather than simply listing action items, it can create tasks and route them to automated agents or to the people named in the meeting. QueueZero positions the product as a step beyond the summary tools that have become common in workplace software over the past two years. The company's chief executive, Evan Wu, was partway through a walkthrough of Proxy's task-assignment features when a customer seated in the audience remarked, "Someone should tell Evan to stop talking and send us the pricing sheet." Proxy, which was running live on the demonstration feed, treated the comment as an instruction. Within moments, two new tasks appeared on screen, both assigned to Wu: "Stop talking" and "Send pricing sheet." Wu laughed, completed the second task by sending the pricing document to attendees and deleted the first. He did not stop talking, which was arguably the correct call for someone still in the middle of a product launch. The exchange, while brief, illustrated the central design question for tools that move from recording meetings to acting on them: how much of what is said should be taken as a directive, and by whom. QueueZero said Proxy is built with that distinction in mind. According to the company, the assistant requires explicit confirmation from a user before it sends any message outside the organization, such as an email to a client or a vendor. Internal tasks, by contrast, can be created automatically based on what Proxy hears, which is what occurred during the demonstration. Administrators have the option to turn off automatic task creation entirely, in which case Proxy will propose tasks for review rather than create them on its own. The company did not indicate that it plans to change the default behavior following the incident. Proxy is priced at $30 per user per month. QueueZero said the product will enter public beta next week, with broader availability to follow. The company has not disclosed how many organizations participated in private testing or how it intends to measure the accuracy of Proxy's task assignments once the system is in wider use. For now, the first documented task Proxy ever assigned remains the one its own CEO chose not to complete.