Head to head: Anthropic: Claude Opus 5.5 vs Qwen: Qwen3.8 2.4T A95B

Anthropic: Claude Opus 5.5: # Raincheck Raises $8 Million to Help Cities Prepare for Floods *The Austin startup's software combines drainage maps with rainfall forecasts to help local governments decide where to send maintenance crews before a storm.* Raincheck, an Austin company that makes flood-risk software for local governments, has raised $8 million in seed funding, the company announced today. The software maps drainage systems and combines those maps with rainfall forecasts. Cities use it to decide where to send maintenance crews before a storm. Basin Ventures led the round, with participation from existing investor Long Mile. The new funding brings Raincheck's total raised to $10.5 million. CEO Ana Flores said the challenge for cities is prioritization rather than identifying flood-prone areas: > "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 [overview of its forecasting methodology](https://example.com/raincheck-methodology). Raincheck works with seven cities in Texas and Louisiana. The company declined to identify them because some contracts have not been publicly announced. Raincheck's forecasts serve as recommendations. Emergency managers retain control over crew deployments and public warnings. The company competes with other flood-planning vendors, including HydroMap, as well as municipal engineering contractors. Raincheck says it will use the new funding to hire eight engineers and expand into Florida.

Qwen: Qwen3.8 2.4T A95B: # Raincheck Raises $8 Million to Help Local Governments Anticipate Flooding *The Austin company’s flood-risk software pairs drainage maps with rainfall forecasts to guide pre-storm maintenance.* Raincheck, a company that makes flood-risk software for local governments, announced today that it has raised $8 million. The Austin company maps drainage systems and combines those maps with rainfall forecasts, giving cities a tool for deciding where to send maintenance crews before a storm arrives. The seed round was led by Basin Ventures, with participation from existing investor Long Mile. Raincheck has now raised $10.5 million in total. CEO Ana Flores said the challenge is not simply knowing where flooding occurs. > “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). 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, and 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 Opus 5.5: # The Cheaper Model Matched the Leader in a Customer-Support Test In a test of how well AI models answer customer-support tickets, the most useful result is a tie, not a winner. The top-scoring model, North, and the runner-up, Vale, finished close enough that researchers could not distinguish them statistically. Vale cost about 61% less to run and answered faster. For teams choosing a model for support work, that matters more than which name sits at the top of the leaderboard. ## How the test worked Researchers gave five language models the same 200 anonymized customer-support tickets. Each model wrote one proposed answer per ticket. Three experienced support managers then graded the answers blindly, without knowing which model wrote which response. Each answer received a combined score covering three qualities: - **Factual correctness:** Was the answer right? - **Usefulness:** Did it help resolve the customer's problem? - **Tone:** Did it sound appropriate for a support interaction? Blind grading by practitioners is a strength of the design. Support managers know what a good answer looks like in practice, and blinding reduces the chance that brand reputation or expectations influenced their scores. ## The headline: a statistical tie at very different prices | 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 | North's 0.4-point lead over Vale was not statistically significant. In plain terms, the data do not show that North produces better answers. A gap that small could come from which tickets happened to be in the sample. The cost difference is much clearer. North was estimated at $1.90 per 1,000 tickets and Vale at $0.74. Vale also answered 18% faster, which can matter in live chat or any setting where customers are waiting. Absolute costs are low for every model in the test. Even North costs under a fifth of a cent per ticket, so for many organizations the price gap will be small next to staffing costs. At high volumes, though, the difference adds up. If quality is effectively equal, there is little reason to pay more than double for it. ## Different strengths, with a caution The two leaders did best on different kinds of tickets. North scored highest on billing disputes, and Vale scored highest on technical troubleshooting. This hints that the right choice may depend on a company's ticket mix. A business dominated by billing questions might lean toward North. A software company fielding mostly technical issues might favor Vale. However, these category results come from subsets of a 200-ticket sample, and the researchers did not report whether the category differences were statistically meaningful. Treat them as leads to check against your own tickets, not settled conclusions. ## The lower tier: a trade-off, not a bargain bin Ember, Kite, and Harbor were cheaper still, with Harbor at just $0.08 per 1,000 tickets. Their scores also dropped as price fell, and Harbor trailed the leaders by roughly 13 points. Whether those models are good enough depends on how they are used. A model that drafts replies for a human agent to review can tolerate more errors than one sending answers directly to customers. The benchmark measured answer quality, not how much editing each draft would need, so it cannot settle that question. ## A shared blind spot: images Every model performed poorly when a ticket depended on an image attachment, such as a screenshot of an error message. This was the clearest weakness in the test, and switching models would not fix it. Teams that receive many screenshot-based tickets should plan to route those to human agents or test image handling separately. ## What the benchmark doesn't tell you The results are useful but narrow: - **One company, one language.** All tickets were in English and came from a single software company. Results may differ for other industries, products, or languages. - **No agentic work.** The models only wrote answers. Researchers did not test tool use or actions like issuing refunds, looking up accounts, or updating records, which many real support deployments require. - **One answer per ticket.** Models can give different answers to the same prompt. A single attempt per ticket doesn't show how consistent each model is. - **A combined score.** Merging correctness, usefulness, and tone into one number can hide important differences. A model that is warm but occasionally wrong is riskier than one that is accurate but a little stiff. ## The takeaway This benchmark doesn't crown a best model. It shows that the top two performed about equally on general support answers, and one did so faster and at well under half the cost. For organizations whose tickets resemble this test set, Vale looks like the stronger value. The sensible next step is to test both leaders on a sample of your own tickets, especially in the categories that matter most to your business, before committing to either.

Qwen: Qwen3.8 2.4T A95B: # The clearest winner in support-ticket benchmarking is value, not the highest score A new benchmark of five language models on customer-support drafting points to a practical conclusion: Model Vale is the most attractive overall option. It scored almost identically to top-scoring Model North, cost substantially less per 1,000 tickets, and generated answers faster. In a task where small score gaps may not reflect real-world quality, that combination is the clearest finding. The benchmark tested 200 anonymized customer-support tickets from one software company. Each model produced one proposed reply for each ticket. Three experienced support managers then graded the answers without knowing which model wrote them. The scoring combined factual correctness, usefulness, and tone — the ingredients that matter most when a support agent is deciding whether a draft can be sent with light edits. Model North earned the highest average score at 84.2, while Model Vale followed at 83.8. That 0.4-point gap is too small to treat as a meaningful win. According to the researchers, the difference was not statistically significant. In other words, based on this test, a support team should not assume North will produce reliably better replies than Vale. The economics make Vale more compelling. North was estimated to cost $1.90 per 1,000 tickets, while Vale cost $0.74 per 1,000 tickets. Vale also answered 18% faster. For high-volume support queues, those differences could affect staffing, response-time targets, and operating cost. A model that is slightly cheaper on paper but much faster at scale may create more value than a marginal quality advantage. This is not to say that the lowest-cost model is always the right answer. The benchmark shows a broad trade-off between price and perceived quality. Ember, Kite, and Harbor were cheaper, but their average scores were lower. Those models might still be useful for internal summaries, triage, or low-risk macros, but the results do not suggest they should replace the leaders for customer-facing drafts. The important point is to avoid reading tiny score differences as decisive. The stronger conclusion is that North and Vale form a top tier, with Vale offering better overall value. The models also had different strengths. North performed best on billing disputes, where precision and careful language can be especially important. Vale performed best on technical troubleshooting, where clear step-by-step guidance matters. A team could reasonably choose based on ticket mix: North may deserve a closer look for billing-heavy workflows, while Vale may be better suited to technical support. Still, across all tickets, Vale’s overall package of quality, price, and speed is easier to justify. One limitation deserves emphasis: every model struggled when a ticket depended on an image attachment. That weakness cuts across all five systems. In real support environments, screenshots, error dialogs, and annotated interface images are common. If a request cannot be understood without an image, these models may produce incomplete or misleading replies. Human review remains essential in those cases, and organizations should not assume text-only benchmark results transfer directly to multimodal ticket queues. The methodology is sensible but narrow. The tickets came from one software company and were in English. Three expert graders reduce the risk of model-name bias, and blind scoring is helpful, but judgments about tone and usefulness may still reflect one company’s support style. The benchmark also evaluated single proposed answers only. It did not test agentic actions, tool use, knowledge-base retrieval, ticket routing, refunds, account changes, or multilingual support. Those capabilities can matter enormously in production systems. The estimated costs are also worth treating as operational approximations rather than universal prices. Actual costs may change with prompt design, context length, caching, volume discounts, or deployment choices. The clearest takeaway is not that Model North “won” by a fraction of a point. It is that Model Vale reached a statistically indistinguishable quality level while costing less than half as much per 1,000 tickets and responding faster. For teams drafting support replies in English, Vale looks like the default starting point. North remains a credible premium option for billing-sensitive work. But for most ordinary tickets, the benchmark suggests that the best model is not the one with the highest score — it is the one that balances quality, cost, and speed well enough to be deployed safely with human oversight.

Anthropic: Claude Opus 5.5: C

Qwen: Qwen3.8 2.4T A95B: C

Anthropic: Claude Opus 5.5: **6** 1. `#Flood software moves upstream` has no space after `#`. 2. `The company is expanding its pilot.` is the dek, but it is not italicized. 3. `##Details` has no space after `##`. 4. The link URL is `www.example.com/news`, which is not an absolute HTTPS URL. 5. `* The first pilot begins Monday.` begins with `*` instead of a hyphen. 6. `<div>The second pilot begins Friday.</div>` is raw HTML.

Qwen: Qwen3.8 2.4T A95B: 6

Anthropic: Claude Opus 5.5: # Foundry Hall to Close After Losing Its Largest Sponsor Foundry Hall, the Milwaukee nonprofit that has helped more than 200 startups get off the ground since 2018, will close October 31 after its largest corporate sponsor declined to renew its support. Blank River Bank had supplied 38% of the accelerator's annual budget. Without that funding, Foundry Hall has enough money to operate through October but not enough to run another full program. That leaves its leaders facing an end they spent months trying to avoid. Executive director Camille Ross said the organization had hoped a new backer would step in to fill the gap. "We kept waiting for the replacement check that never came," she said. The closure will eliminate six jobs. It will also displace the 12 startups now working out of Foundry Hall's space. Those companies must find somewhere else to go before the doors close, a difficult task for young businesses that often chose the accelerator because they could not yet afford offices of their own. A spokesperson for Blank River Bank said the decision reflected a shift in the bank's community investment priorities and had nothing to do with how Foundry Hall had performed. By the measures the accelerator tracks, it had performed well. The 212 startups it has supported now collectively employ 1,140 people, a footprint that will remain in Milwaukee's economy after the organization that nurtured it is gone. For some founders, the loss is more personal than statistical. Dev Malik, who went through the program, remembers what his company looked like before he arrived. "My first investor meeting happened in that building," Malik said. "Before Foundry Hall, I was pitching from the back table of a coffee shop." That kind of starting point, a desk, an address, and a room where an investor might take a founder seriously, is what the 12 current companies stand to lose. City officials are trying to soften the landing. Deputy Mayor Susan Cole said the city is in talks with building owners and local universities about finding space for those businesses. For now, those conversations offer a possibility rather than a guarantee. Foundry Hall's closing comes not from failure but from a funding model that leaned heavily on a single sponsor. When nearly two-fifths of the budget disappeared, there was little margin left to absorb the blow. In its final weeks, the accelerator will do what it has done for six years: give early-stage companies a place to work while they figure out whether their ideas can survive. This time, the companies will have to keep figuring it out somewhere else. The building's doors will close on October 31. What happened inside them, measured in companies, jobs, and first meetings that led somewhere, will carry on without it.

Qwen: Qwen3.8 2.4T A95B: # The Last Lights in the Foundry Foundry Hall will close on October 31, ending a six-year run in which the nonprofit startup accelerator became a first address for young companies in Milwaukee. Since opening in 2018, it has supported 212 startups that together employ 1,140 people, a measure of how much of the city’s new economy passed through its rooms. The closing is not the result of an empty calendar or a failed experiment. It is a budget hole. Blank River Bank, Foundry Hall’s largest corporate sponsor, supplied 38% of its annual budget and declined to renew that support. The bank said its decision reflected a change in community investment priorities and was unrelated to the accelerator’s performance. Without that money, Foundry Hall has enough money to operate through October, but not enough to carry another full program. That leaves a small staff facing the end of a project they helped build. Six employees will lose their jobs. Twelve startups currently using the workspace must leave, scattering companies that had been sharing desks, advice, and the daily reassurance that other people are trying to do the same hard thing. Executive director Camille Ross described the final months as a long wait for a rescue that never arrived. “We kept waiting for the replacement check that never came,” she said. For founders, the building was more than square footage. Dev Malik, whose company went through the program, remembered Foundry Hall as the place where he could finally be taken seriously. His first investor meeting happened there, after a period of pitching from the back table of a coffee shop. The city is trying to soften the landing. Deputy mayor Susan Cole said officials are speaking with the building’s owners and local universities about space for the current companies. Those discussions may help some startups survive the transition, but they cannot replace the ecosystem that made Foundry Hall useful in the first place. The accelerator’s closure is a reminder that nonprofit innovation spaces often rest on fragile funding. They are asked to produce jobs, momentum, and civic pride, yet their survival can depend on a single sponsor’s renewal. Foundry Hall managed to do the visible work — companies formed, founders trained, payrolls created — while carrying an invisible risk: that one major commitment could disappear. When the doors close at the end of October, the city loses more than a venue. It loses a room where uncertain ideas had somewhere to become companies, and where people who were just beginning could look around and realize they were not alone.

Anthropic: Claude Opus 5.5: # Startup Spotlight: LedgerLark Wants AI Agents to Handle the Back-Office Grind After a trade is executed, a quieter process begins. Confirmations arrive, custodians update their records, and internal ledgers have to agree with both. When they don't, someone has to figure out why. At small investment firms, that someone is often a stretched operations team working through spreadsheets and email threads. LedgerLark, a young startup, is betting that AI agents can take on a meaningful share of that work. ## A three-day problem with a one-line cause The idea traces back to co-founder Mina Batra's seven years in operations at a regional brokerage. There, she watched a six-person team spend three days untangling a single trade mismatch. The cause turned out to be a copied account number. The episode was minor on its own, but it pointed to something broader. Post-trade errors are often small, mechanical and tedious to locate, and the cost of finding them is measured in skilled people's time. Batra began exploring whether software could do more of the searching. Her co-founder, Owen Carr, brings a complementary background. He built reconciliation systems at a large bank, giving him experience with how institutions match records at scale and, presumably, with where those systems fall short. ## What the product does and doesn't do LedgerLark's agents compare three sources of truth: trade confirmations, custodian records and a firm's internal ledgers. When they find discrepancies, they propose corrections. A human then approves or rejects each one. That boundary matters. The product does not execute trades or move money. In a field where a mistaken correction can ripple into regulatory and client problems, keeping people in the approval seat may make the software easier for cautious firms to adopt. It also means the product's value rests on how well the agents investigate and explain, not on full automation. An open question is how the agents perform on messier cases, such as ambiguous data, unusual instruments or errors that span multiple systems. Proposing a fix for a copied account number is one thing. Diagnosing a chain of related breaks is another. ## Early signs LedgerLark says five firms are paying customers, with annual contracts ranging from $36,000 to $110,000. The company also says one customer reduced its unresolved daily exceptions from 70 to 18. That figure, if it holds up, suggests real time savings. But it comes from the company and describes a single customer. On its own, it doesn't reveal how the change was measured, how long it took or whether other factors played a role. A fuller picture will depend on results across more firms and over longer periods. The company has raised a $3.5 million seed round led by Motive Row. ## A crowded, careful market LedgerLark is not entering empty territory. Established reconciliation vendors already sell into this space, and large banks often rely on internal software. The startup's focus on small investment firms may help it avoid competing head-on with the biggest institutions. But smaller firms also have smaller budgets and may be content with existing tools or manual processes. Sales cycles are another hurdle. Financial institutions may require lengthy security and procurement reviews before granting a vendor access to sensitive records. For a seed-stage company, those timelines can strain resources and slow growth regardless of how well the product works. ## What to watch The core insight is easy to grasp: many costly operational headaches begin with small errors that are hard to find. LedgerLark now has to show that AI agents can find those errors reliably, explain their reasoning clearly and earn the trust of risk-averse operations teams. It will have to do that one customer, and one review process, at a time.

Qwen: Qwen3.8 2.4T A95B: # Startup Spotlight: LedgerLark Looks at the Quiet Work After a Trade The trade itself often gets the attention, but the work that follows can be surprisingly manual. LedgerLark is building AI agents for post-trade operations at small investment firms, starting with the task of spotting and resolving mismatches between records. The idea came from co-founder Mina Batra’s time inside operations. She spent seven years at a regional brokerage, where she saw how a small error could expand into days of work. In one case, a copied account number caused a trade mismatch that took a six-person team three days to resolve. That experience pointed to a practical problem: firms may not need more dashboards, but help turning scattered exceptions into a manageable workflow. Batra founded LedgerLark with Owen Carr, who previously built reconciliation systems at a large bank. Their combined backgrounds suggest a product shaped by both front-line operations and back-end systems work, rather than by a broad ambition to “disrupt” finance. LedgerLark’s agents compare confirmations, custodian records and internal ledgers, then propose corrections for human approval. The company emphasizes that the product does not execute trades or move money. That distinction matters in financial services, where firms are likely to be cautious about letting software take actions that affect client assets or regulatory records. By proposing changes instead of making them automatically, LedgerLark keeps a person in the loop while aiming to reduce the time spent hunting for discrepancies. The company’s early traction is modest but specific. It has raised a $3.5 million seed round led by Motive Row, and has five paying customers, with annual contracts ranging from $36,000 to $110,000. It also says one customer reduced unresolved daily exceptions from 70 to 18. That figure is worth noting, though it comes from the company, and a single customer result may not predict broader outcomes across different firms, asset classes or data environments. Still, the problem appears concrete enough for a narrow wedge. Small investment firms may not have the engineering teams or budgets of large banks, yet they still face the same basic pressure: records must match, breaks must be investigated, and unresolved exceptions can create operational and compliance risk. If LedgerLark can reduce routine investigation work, it may give operations staff more time to handle cases that actually require judgment. The market, however, is not empty. LedgerLark competes with established reconciliation vendors and with internal software built by banks and other financial institutions. Those incumbents may already have relationships, security approvals and integrations that a young startup must match. In addition, financial institutions can require lengthy security and procurement reviews, which may slow sales cycles and make early growth more uneven than a simple product demo suggests. That uncertainty is part of what makes the company interesting. LedgerLark is not claiming to replace operations teams or automate the entire back office. Its pitch is narrower: use AI to compare records, surface likely fixes and let people approve them. Whether that approach can scale beyond five customers, survive rigorous vendor reviews and compete with larger systems remains an open question. For now, LedgerLark offers a useful example of how AI is moving into less glamorous parts of financial operations. The opportunity may be less about speed alone and more about making back-office work easier to see, sort and resolve.

Anthropic: Claude Opus 5.5: # KiteBridge Software Glitch Sends $2.8 Million in Duplicate Reimbursements A faulty software update at expense-management company KiteBridge sent out $2.8 million in duplicate employee reimbursements on Tuesday, leaving 91 corporate customers to cover payments they never meant to make. More than a day later, about $500,000 of that money had still not been recovered. The errant payments, 13,400 in all, went out over roughly an hour on Tuesday morning before KiteBridge halted its automated payment system. The company's public statements about the problem drew scrutiny of their own. For several hours, KiteBridge described the incident only as a delay in reimbursements. ## How the failure unfolded KiteBridge sells software that companies use to track employee expenses and pay back workers for costs such as travel and meals. When an employee's expense is approved, the system sends a payment instruction to a payment processor, the outside service that actually moves the money. The trouble began after KiteBridge deployed version 4.2 of its software at 8:45 a.m. Tuesday. A payment processor experienced a timeout, meaning it failed to respond within the expected window. In that situation, software typically "retries," resending the same request to make sure it goes through. KiteBridge's fallback system, a backup mechanism designed to keep payments flowing when something goes wrong, mishandled those retries. Instead of recognizing them as repeat attempts at the same payment, it treated each one as a new reimbursement request. Each retry produced another payment. The first customer complaint reached KiteBridge at 9:12 a.m., 27 minutes after the update went live. The company suspended automated payments at 10:03 a.m., about 50 minutes after that first complaint and nearly 80 minutes after the deployment. ## Who paid The duplicate payments were funded by the employers who use KiteBridge, not by their workers. No employees had money withdrawn from their accounts as a result of the error. Instead, employees in effect received extra reimbursements drawn from their companies' funds. By Wednesday evening, $2.3 million of the $2.8 million had been recovered. The remaining money was still pending. ## A promise of protection The incident has raised questions about a feature KiteBridge previously advertised. In April, the company's changelog, a public record of software updates, said version 4.0 had introduced "exactly-once payment protection." In payments software, "exactly-once" refers to a guarantee that each transaction is carried out a single time, even when requests are retried or systems fail partway through. Preventing duplicate charges and payouts is precisely what such protection is meant to do. KiteBridge has not yet explained why the safeguard did not stop Tuesday's duplicates or how the version 4.2 update related to it. ## Slow disclosure KiteBridge's public incident page, where the company posts updates on service problems, initially described the issue as delayed reimbursements. That description was accurate for employees awaiting payment after automation was switched off, but it did not mention that thousands of duplicate payments had already gone out. The page first disclosed the duplicate payments around 4 p.m. Tuesday, six hours after the company suspended automated payments. Chief Executive Lena Ortiz apologized for the incident and said KiteBridge would publish a technical report. The company did not say when the report would be released.

Qwen: Qwen3.8 2.4T A95B: # KiteBridge Update Created $2.8 Million in Duplicate Reimbursements A routine software update at expense-management software maker KiteBridge produced 13,400 duplicate reimbursements worth $2.8 million on Tuesday, affecting 91 corporate customers and forcing the company to suspend automated payments. The incident began at 8:45 a.m. Tuesday, when KiteBridge released version 4.2 of its platform. Within half an hour, the first customer complaint arrived, signaling that something had gone wrong in the reimbursement pipeline. By mid-morning, KiteBridge had halted automated payments, but not before the error had sent extra funds to employees across customer organizations. The failure stemmed from a timeout involving an outside payment processor. In such cases, KiteBridge’s fallback system is meant to preserve or retry payment requests so reimbursements are not lost. Instead, the fallback treated retried requests as new reimbursement requests. That allowed the same legitimate expense payment to be processed more than once. For most users, expense software is invisible infrastructure: employees submit receipts, companies approve claims, and payments move automatically. The duplicate reimbursements therefore did not take money from employees’ bank accounts. Instead, the extra payments were funded by employers, creating unexpected outlays and a recovery effort for both customers and KiteBridge. KiteBridge suspended automated payments at 10:03 a.m. Tuesday. The company’s public incident page initially described the issue as delayed reimbursements. It did not disclose that payments had been duplicated until six hours after the automation was shut down, leaving some customers without a clear explanation of why extra transactions had appeared. By Wednesday evening, KiteBridge said it had recovered $2.3 million of the duplicated funds. The remaining balance was still pending, and the company did not specify when all recovery work would be completed. The episode is likely to raise questions about KiteBridge’s safeguards. In April, the company’s changelog said version 4.0 had introduced “exactly-once payment protection,” a feature intended to prevent a single reimbursement from being paid multiple times even if a payment step fails or must be retried. Version 4.2 was the latest deployment, but the company has not yet explained why that protection failed during the update or whether it was bypassed by the fallback logic. KiteBridge is a vendor that manages employee expense claims and reimbursement payments for corporate clients, so a single technical defect can ripple across many finance teams at once. Ninety-one companies were affected, meaning finance departments may need to reconcile payroll, expense, and cash-flow records. Chief Executive Lena Ortiz apologized and said KiteBridge would publish a technical report. That report will need to explain how a payment-processor timeout became a large-scale duplication event, why the incident page first characterized the issue as a delay, and what changes will prevent similar failures in future releases. Until then, customers are left waiting for the final recovery of roughly $500,000 and for assurance that KiteBridge’s promise of exactly-once payments can be restored.

Anthropic: Claude Opus 5.5: # HelioDock Launches Night Shift to Let Operators Oversee Warehouse Robots Remotely *The autonomous robot maker also raised a $24 million Series B round. The software will cost $90 per robot per month when it becomes generally available Sept. 15.* HelioDock, a maker of autonomous warehouse robots, on Thursday launched Night Shift, software that lets operators monitor their robot fleets remotely. The company also announced a $24 million Series B funding round led by North Coast Ventures. Night Shift is designed to handle small problems that can stop a robot and, without remote access, bring a manager to the site in person. HelioDock says customers who tested the software saw fewer unplanned stoppages, but that figure has not been independently verified. At least one early user says the system's usefulness depends heavily on a facility's network connection. The product also enters a market where Relay Robotics and YardOS already sell fleet-management tools. ## The problem HelioDock is targeting Chief Executive Mara Velez described the product in terms of a scenario many warehouse operators will recognize. "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. Her example points to a particular cost of automation. A robot that stalls on an ordinary obstacle can stop work until someone intervenes, even when the fix is minor. Night Shift is meant to let that person deal with the problem from a distance. ## Results from a 14-month beta HelioDock tested Night Shift for 14 months in a private beta involving nine logistics companies and 310 robots. According to the company, beta customers had 17% fewer unplanned robot stoppages. The company did not provide independent verification of that number. Because it comes from HelioDock's own reporting on a limited group of customers, it is best read as an early, company-supplied indication of how the software performed, not as an established benchmark. ## Where the connection drops Night Shift's remote features require a working network link. When connectivity fails, HelioDock's robots return to their locally configured behavior, and remote control becomes unavailable. Aaron Pike of Fulcrum Logistics, one of the beta customers, said the system worked well in facilities with strong connectivity. During LTE outages at rural sites, he said, it became less useful. That fallback means robots are not left waiting for instructions when the link goes down. They revert to the settings configured on site. It also means the remote oversight that Night Shift offers can disappear at the moment an operator may want it. Companies whose warehouses depend on cellular connections in rural areas may get less from the software than those with more reliable networks. ## Pricing and competition Night Shift will become generally available Sept. 15 at $90 per robot per month. Because pricing is per unit, costs rise directly with fleet size. A 50-robot deployment would cost $4,500 a month, or $54,000 a year. A fleet the size of the entire beta, 310 robots, would cost $27,900 a month. Buyers will be able to compare Night Shift with fleet-management products from Relay Robotics and YardOS. HelioDock's results claim and its connectivity limits are likely to be among the factors operators weigh as they evaluate the options. ## New funding The Series B brings HelioDock's total funding to $39 million, which means the company had raised $15 million before this round. North Coast Ventures led the round.

Qwen: Qwen3.8 2.4T A95B: # HelioDock Launches Night Shift Remote Monitoring for Warehouse Robots *The software, priced at $90 per robot monthly, follows a 14-month beta with nine logistics companies and arrives alongside a $24 million Series B.* HelioDock on Thursday launched Night Shift, software that allows warehouse operators to monitor fleets of autonomous robots remotely, as the company seeks to reduce the after-hours disruptions that can arise when machines encounter problems without nearby staff. The launch extends HelioDock’s business beyond making autonomous warehouse robots and into the software used to manage them. The company said beta customers experienced 17% fewer unplanned robot stoppages, a claim it supplied that has not been independently verified. The release also comes with funding: a $24 million Series B led by North Coast Ventures that brings HelioDock’s total raised to $39 million. Night Shift spent 14 months in private beta involving nine logistics companies and 310 robots. HelioDock said the software will become generally available on Sept. 15 and will cost $90 per robot per month. That pricing links the software’s cost to the size of a customer’s fleet. The product is aimed at a routine problem in automated warehouses: robots can stop or require attention while human staff are not on site. CEO Mara Velez said the software is intended to prevent managers from having to respond personally to those incidents. “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. The 17% reduction claim is the main performance figure HelioDock has offered for the beta period. Because the number is company supplied and has not been independently verified, it does not yet show how consistently the software performed across different facilities or network conditions during the 14-month test. During the beta, nine logistics companies used Night Shift with 310 robots over more than a year. That testing window gave HelioDock customer feedback ahead of general availability, though the company’s performance claim has not been independently verified. Fulcrum Logistics’ Aaron Pike, a beta customer, offered a mixed assessment. He said the system worked well in facilities with strong connectivity but became less useful during rural LTE outages. When connectivity fails, HelioDock’s robots return to their locally configured behavior, and remote control becomes unavailable. That limitation affects how much a facility can rely on remote operations. In a warehouse with dependable wireless service, Night Shift can provide oversight without requiring a person to be physically present at the moment a robot stops. In a rural site or during an LTE outage, the software cannot provide remote control, and the robots operate according to local settings. Pike’s comments also show that the product’s value may differ between facilities with stable connectivity and those with less reliable LTE service. A facility with strong connectivity can have operators monitor robots from a distance. A facility that experiences LTE outages may still need local procedures to handle robots while remote access is down. The monthly fee means operators will pay for Night Shift according to the number of robots in a fleet. A larger deployment carries a larger recurring software cost, while a smaller fleet pays for fewer monitored robots. HelioDock is entering a category where other vendors already sell fleet-management software. Competing products are available from Relay Robotics and YardOS. The presence of those alternatives gives logistics companies options beyond HelioDock’s own robots and software when they evaluate remote monitoring tools. North Coast Ventures led the $24 million Series B, bringing HelioDock’s disclosed total to $39 million. Night Shift is scheduled to become generally available Sept. 15.

Anthropic: Claude Opus 5.5: # Cloudnote Debugging Page Exposed Prompt Excerpts From 1,842 User Sessions *The page was reachable without a login for four hours Monday. The company says it has found no sign that anyone but the researcher who reported it viewed the data, though its investigation is not finished.* A Cloudnote debugging page containing excerpts of customer prompts was accessible without authentication for four hours on Monday, the company said. Logs show the page held text from 1,842 user sessions. Each excerpt could include up to 200 characters of a prompt, a fragment rather than a full submission but potentially enough to capture whatever a user had typed at the start of a request. Cloudnote said full documents, passwords and payment information are stored separately and did not appear on the page. The company removed the page 19 minutes after a researcher notified it of the problem. Cloudnote says it has found no evidence that anyone other than the researcher accessed the page. It has not completed its investigation, and key questions remain open: whether anyone else viewed or copied the excerpts, and what the exposed text contained. Because the figure counts sessions, it is also not yet clear how many individual users were affected. The sessions involved represent a small share of Cloudnote's 3.2 million registered users. But the incident shows how internal tools built for troubleshooting can expose material that users had reason to believe was private, even when a company's most sensitive data is kept elsewhere.

Qwen: Qwen3.8 2.4T A95B: # Cloudnote Debugging Page Exposed Customer Prompt Excerpts for Four Hours *The company says 1,842 user sessions were logged during the exposure, and the investigation is continuing.* Cloudnote, a note-taking service with 3.2 million registered users, said Monday that a debugging page containing excerpts of customer prompts was accessible without authentication for four hours. The company confirmed that the page was removed 19 minutes after a security researcher reported it. According to Cloudnote, logs show excerpts from 1,842 user sessions appeared on the page. Each excerpt could include up to 200 characters from a prompt. The company has not determined whether any of those excerpts contained sensitive information, and it has not said which users may have been affected. Cloudnote said it has found no evidence that anyone other than the researcher accessed the page, but its investigation is not complete. Full documents, passwords and payment information were stored elsewhere and were not present on the exposed page, the company said. The incident matters because even limited prompt fragments can reveal personal details, work topics or private intentions, depending on what users type into an AI assistant. It also highlights how internal debugging tools, often built for engineers rather than end users, can create unexpected privacy risks if they are left reachable on the public internet.

Anthropic: Claude Opus 5.5: # 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 2.4T A95B: # 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 Opus 5.5: # Arcwell's Relay Puts AI Suggestions in Front of Customer Support Agents Arcwell is betting that the answer to most customer questions already exists somewhere inside a company, and that support agents shouldn't have to dig for it. The company has announced Relay, an artificial intelligence platform that reads customer support conversations and suggests responses to the agents handling them. Beyond drafting answers, Relay can flag customers who may be about to cancel and can generate summaries. "Support agents spend too much time searching for information that their companies already have," said Priya Shah, Arcwell's chief executive. Arcwell tested Relay with 12 companies over six months. According to the company, agents using the platform closed tickets 14% faster. That figure comes from Arcwell itself, and the trial's results are incomplete. The company did not release customer-retention results, so it remains unclear how well Relay's cancellation warnings performed in practice. For buyers, the churn feature may prove as important as the speed gains, since faster tickets matter less if customers leave anyway. Relay costs $65 per agent each month, a price that scales directly with the size of a support team. The suggestions depend on access. Relay needs a company's support transcripts and its internal help documents, which serve as the material it draws on to recommend answers. That gives businesses a familiar trade-off: quicker resolutions in exchange for handing customer conversations and internal knowledge to an outside vendor. Arcwell has moved to address that concern. It says customer data is encrypted and is not used to train shared models, so one client's transcripts should not shape the suggestions another client receives. Companies also control how long their data is kept, choosing a retention period of either 30 or 90 days. Those safeguards answer some of the questions a security team is likely to ask. Others, including whether Relay keeps customers from leaving, will have to wait until Arcwell shares more of what its testers saw.

Qwen: Qwen3.8 2.4T A95B: # Arcwell Launches Relay, an AI Platform for Customer Support Agents Arcwell has announced Relay, an artificial intelligence platform that examines customer support conversations and suggests answers to agents. The launch expands Arcwell’s role in the broader customer service industry. The product brings together response assistance, cancellation signals and conversation summaries in a single tool aimed at improving customer service operations. Relay works by analyzing support exchanges and company knowledge sources. It can suggest answers to agents, identify customers who may cancel and create summaries of interactions. Those capabilities show how artificial intelligence continues to change customer service work, offering help during conversations while producing records that teams can review later. CEO Priya Shah said the platform addresses a familiar problem. “Support agents spend too much time searching for information that their companies already have,” Shah said. Relay is intended to reduce that search by using information already contained in support transcripts and internal help documents. Arcwell tested the platform with 12 companies over six months. The company said agents using Relay closed tickets 14% faster, making that figure the main performance result disclosed from the trial. It did not release customer-retention results, so the effect of the cancellation-warning feature remains undisclosed. Relay costs $65 per agent each month. Because the platform requires access to support transcripts and internal help documents, data handling is central to its design. Arcwell said customer data is encrypted and is not used to train shared models. Companies can choose a 30-day or 90-day retention period. The launch positions Arcwell as a provider of AI assistance for companies seeking to improve customer service. Relay’s combination of suggested answers, churn signals and automated summaries gives support teams several related functions in one system, while its privacy and retention controls offer buyers a measure of oversight. The result is a product that ties productivity, customer insight and data governance together for support managers evaluating it.

Anthropic: Claude Opus 5.5: # QueueZero's AI Meeting Assistant Takes a Customer's Aside Literally at Launch **Proxy, which can attend meetings and hand off follow-up work to software agents, enters public beta next week at $30 per user per month.** QueueZero introduced Proxy, an AI meeting assistant, on Wednesday, pitching it as a way to offload the administrative work that follows a meeting. The software can attend a meeting, produce notes, and assign follow-up tasks to software agents. The launch demonstration also offered an unscripted look at how the system decides what counts as an instruction. During the demonstration, a customer in the audience said, "Someone should tell Evan to stop talking and send us the pricing sheet." Proxy treated the remark as a directive. It created two tasks, both assigned to QueueZero CEO Evan Wu: "Stop talking" and "Send pricing sheet." Wu laughed, completed the second task, and deleted the first. ## Commands versus conversation The moment was lighthearted, but it points to a practical question for any tool that listens to meetings and acts on what it hears: how reliably it can separate a genuine request from a passing comment, a joke, or a hypothetical. Meetings are full of phrases like "someone should" and "we ought to," and not all of them are meant to produce work. QueueZero says Proxy draws a line between internal and external actions. According to the company, the assistant requires confirmation before an external message is sent. Internal tasks, by contrast, can be created automatically, which is how the two items appeared during the demonstration without anyone approving them. Administrators can disable automatic task creation, according to the company. That gives organizations a choice between speed and control. They can allow Proxy to populate task lists on its own, or they can require a person to review its proposals before anything is assigned. For teams evaluating the product, the distinction matters. An unwanted task on an internal list is easy to delete, as Wu showed. A message sent to a client or partner because of a misread remark would be harder to take back. QueueZero's confirmation requirement is designed to prevent that scenario. ## Pricing and availability Proxy costs $30 per user per month. It enters public beta next week. The beta period will give QueueZero a broader test of how Proxy interprets real conversations outside a controlled launch event. By the end of Wednesday's demonstration, at least, the customer had the pricing sheet.

Qwen: Qwen3.8 2.4T A95B: # QueueZero’s AI Meeting Assistant Creates Tasks From Audience Remark During Launch QueueZero launched an artificial-intelligence meeting assistant called Proxy on Wednesday, positioning the product as a way for workers to delegate meetings and the follow-up work that comes with them. The company demonstrated Proxy’s ability to attend a meeting, generate notes and assign subsequent tasks to software agents. During the demonstration, a customer in the audience said, “Someone should tell Evan to stop talking and send us the pricing sheet.” Proxy interpreted the comment as an instruction rather than an aside. It created two tasks assigned to QueueZero chief executive Evan Wu: “Stop talking” and “Send pricing sheet.” Wu laughed, completed the second task and deleted the first. The moment gave the launch an unplanned test case. It also showed how quickly AI assistants can move from listening to acting, particularly when they are trained to identify action items in ordinary conversation. QueueZero said Proxy requires confirmation before an external message is sent. Internal tasks, however, can be created automatically, depending on how the product is configured. Administrators can disable automatic task creation, a control that is likely to be important for companies testing AI agents in busy workplace environments. The incident highlights a broader issue facing makers of workplace AI tools. As assistants become more capable, they must distinguish between remarks, suggestions and commands, especially in meetings where people speak casually. A sentence that is clear to humans may be ambiguous to software. QueueZero is entering a crowded market for AI meeting helpers. Products from established software companies and startups already offer transcription, summaries and automated follow-ups. QueueZero is betting that its agents can do more than record what happened, by turning discussion into tasks that software can carry out or assign. Proxy will cost $30 per user per month and enters public beta next week. The pricing places it in the range of premium productivity tools, suggesting QueueZero is aiming at businesses that want to reduce time spent on meetings and coordination. For QueueZero, the launch incident may become part of the product’s story. The company showed that Proxy can listen, interpret and act. It also showed why human oversight remains a necessary feature.