Head to head: MoonshotAI: Kimi K3.1 Pro vs Anthropic: Claude Fable 5.1

MoonshotAI: Kimi K3.1 Pro: # Arcwell launches Relay, an AI platform for customer support teams Arcwell has announced Relay, an artificial intelligence platform that examines customer support conversations and suggests answers to agents. The company says agents using the software closed tickets 14% faster during a six-month trial involving 12 companies. Relay can also create summaries and identify customers who may cancel. Those features extend its role beyond drafting responses: agents can receive help during a conversation, while summaries provide a condensed record and cancellation warnings highlight customers who may need attention. Arcwell is presenting the three capabilities as one system for companies seeking to improve customer service with AI. CEO Priya Shah said the platform addresses the time agents lose hunting for material within their own organizations. “Support agents spend too much time searching for information that their companies already have,” she said. The reported 14% gain measures how quickly agents closed tickets. It does not, on its own, show whether Relay improved customer retention, and Arcwell did not release customer-retention results. The announcement therefore provides a company-reported productivity figure, but no published outcome for the cancellation-related feature. Relay costs $65 per agent each month. The price scales with the number of agents using the platform, making the size of a support team central to the cost of deployment. To generate suggestions, summaries and warnings, Relay requires access to support transcripts and internal help documents. The requirement connects Shah’s diagnosis to the product’s design: to surface information spread across those sources, Relay needs companies to make the sources available. That gives it both records of customer conversations and the material agents use to answer questions. It also makes data handling a central consideration for prospective customers. Arcwell says customer data is encrypted and is not used to train shared models. Companies can choose between 30-day and 90-day retention periods, the two options Arcwell offers for how long information is kept.

Anthropic: Claude Fable 5.1: # Arcwell's Relay Puts an AI Co-Pilot Beside the Support Desk Arcwell has introduced Relay, a platform that reads customer support conversations and hands agents suggested replies while they work. The system also flags customers who appear likely to cancel and produces summaries of each interaction. The company's pitch is speed. Twelve companies tested Relay over six months, and Arcwell says agents using it closed tickets 14% faster than before. That is the only hard number the company has shared. Arcwell did not release customer-retention results, so there is no public evidence yet on whether the cancellation alerts translate into fewer lost accounts — arguably the feature with the most direct bearing on revenue. Relay costs $65 per agent per month. Chief executive Priya Shah framed the product as a fix for an old problem rather than a new capability. "Support agents spend too much time searching for information that their companies already have," she said. That framing points to how Relay works, and to the question buyers will have to settle before signing on. To surface the right answer, the platform needs access to a company's support transcripts and internal help documents — the same material that tends to be most sensitive and most scattered. Arcwell says customer data is encrypted and is not used to train models shared across its clients. Companies can choose whether Relay retains that data for 30 days or 90 days. Those assurances address the obvious worry: that one firm's support history might quietly improve a competitor's bot. They do less to answer how the suggestions perform when the underlying documentation is thin or outdated, a common condition in support organisations and one the pilot results do not speak to. For now, Relay arrives with a modest, measurable claim about ticket times, a clear price, and a retention story that remains to be told. Whether the 14% gain holds up outside a six-month pilot, and whether the churn predictions prove accurate enough to act on, will decide if Relay becomes a fixture on the agent's screen or another tool that gets switched off after the trial period ends.

MoonshotAI: Kimi K3.1 Pro: # Startup Spotlight: LedgerLark and the work after a trade LedgerLark is building AI agents for post-trade operations at small investment firms. Its focus is the investigative work that begins when records disagree: comparing confirmations, custodian records and internal ledgers, then proposing corrections for a person to approve. The product does not execute trades or move money. That boundary is central to the company’s pitch. Rather than asking a firm to hand over control of transactions, LedgerLark is asking whether software can assemble evidence and shorten the path to a reviewed decision. ## An operational problem, seen up close Co-founder Mina Batra spent seven years in operations at a regional brokerage. She began exploring the idea after watching a six-person team spend three days resolving a trade mismatch caused by a copied account number. The episode offered a specific product question: how much of that time went into locating and comparing information, rather than making the final judgment? Co-founder Owen Carr, who built reconciliation systems at a large bank, brings experience with the underlying record-matching problem. The founders’ insight is not that every exception can be automated. It is that some of the repetitive work around an exception may be separable from approval. That distinction leaves room for human review while giving LedgerLark a concrete task to measure: whether its agents can reduce the effort needed to reach an accurate proposed correction. Human approval is a safeguard, but it is not a measure of product quality by itself. A suggestion that is difficult to verify could simply shift work to reviewers. LedgerLark will need to show that its proposals are both checkable and useful, particularly when records are incomplete or an exception falls outside a familiar pattern. ## Early customers, limited evidence Five firms are paying customers, with annual contracts ranging from $36,000 to $110,000. Those contracts indicate willingness to pay, but five customers are not enough to show how consistently the product performs across different systems, exception types and operating procedures. LedgerLark says one customer reduced its unresolved daily exceptions from 70 to 18. That is a useful early data point, not yet evidence of a typical outcome. The figure would be more informative with detail on the measurement period, the customer’s starting process and whether the improvement persisted. It also does not show how much review time the agents save or how often their proposals need to be changed. The company has raised a $3.5 million seed round led by Motive Row. The financing arrives while the commercial and operational questions are still open. ## A crowded path to adoption LedgerLark competes with established reconciliation vendors and with software built inside banks. It must therefore demonstrate more than an ability to compare records. Prospective customers will need a reason to add or switch systems, and a clear account of how the agents handle unfamiliar cases, errors and auditability. Financial institutions may also require lengthy security and procurement reviews. Those processes could slow adoption even when a product is useful, making customer acquisition an important test alongside technical performance. LedgerLark’s narrow scope makes its progress easier to assess. The next useful evidence would be customer results measured over time, including exception resolution, reviewer workload and the quality of proposed corrections. For now, the company has identified a specific operational problem and signed a small group of paying customers. Broader, repeatable results would show more clearly whether its approach can earn a place alongside established tools.

Anthropic: Claude Fable 5.1: # Startup Spotlight: LedgerLark Wants to Clean Up the Back Office, One Mismatch at a Time Most of what happens after a trade is executed is invisible to the people who placed it. Confirmations arrive, custodians report positions, internal ledgers update, and someone has to make sure all three agree. At small investment firms, that someone is often a handful of operations staff working through spreadsheets and email. LedgerLark, a startup building AI agents for post-trade operations, is betting that this work can be partially automated without handing software the keys to anything that matters. ## The problem the founders saw up close Co-founder Mina Batra spent seven years in operations at a regional brokerage. She says the idea took shape after she watched a six-person team spend three days tracing a trade mismatch that turned out to be caused by a copied account number. The error was trivial; the cost of finding it was not. Her co-founder, Owen Carr, built reconciliation systems at a large bank. That experience cuts both ways. He knows how the established tooling works, and he also knows why large institutions have the budget and staff to run it while smaller firms frequently do not. The two founders' backgrounds suggest a fairly specific thesis: the technical problem of matching records is well understood, but the labor of investigating exceptions is still largely manual, and that labor is most painful at firms too small to build their own fix. ## What the product actually does LedgerLark's agents compare trade confirmations, custodian records and a firm's internal ledgers. When they find a discrepancy, they propose a correction and route it to a human for approval. The system does not execute trades and does not move money. That boundary is a design choice worth noting. A lot of the anxiety around AI in financial services concerns autonomous action. By keeping the agent in a propose-and-wait role, LedgerLark sidesteps the hardest version of that conversation. It also means the product's value depends on how good its proposals are. If operations staff end up second-guessing most of them, the time savings shrink. ## Early traction, with caveats Five firms are paying customers, with annual contracts ranging from $36,000 to $110,000. The company says one customer reduced its unresolved daily exceptions from 70 to 18. That figure comes from LedgerLark rather than from an independent review, and a single customer's result is not a pattern. It is also unclear from the company's account how much of the reduction came from the software versus process changes made alongside it. Still, the number gives a sense of the scale the founders are aiming at: not eliminating exceptions, but shrinking the pile to something a small team can clear in a morning. LedgerLark has raised a $3.5 million seed round led by Motive Row. ## The competition and the friction The company is not entering an empty field. Established reconciliation vendors have long sold to this market, and larger banks run internal software that does much of the same matching. LedgerLark's differentiation rests on the investigation layer, the part where a person normally reads through records and figures out what went wrong. Whether incumbents can bolt similar capabilities onto their existing products is an open question, and they have the advantage of already being inside the building. Getting inside the building is the other challenge. Financial institutions, even small ones, may require lengthy security and procurement reviews before adopting new software that touches trade data. For a startup with five customers and a seed round, long sales cycles are a real constraint on how fast it can learn what works. ## What to watch The interesting test for LedgerLark is not whether its agents can find mismatches. It is whether operations staff come to trust the proposals enough to approve them quickly, and whether that trust holds up when the discrepancies get strange. The next few customers will say more about that than the first five.

MoonshotAI: Kimi K3.1 Pro: # QueueZero launches Proxy, an AI assistant that turns meetings into assigned work QueueZero on Wednesday launched Proxy, an AI meeting assistant that can attend meetings, produce notes and assign follow-up work to software agents. The product enters public beta next week and will cost $30 per user per month. Its launch demonstration also illustrated a challenge facing this category of workplace software: determining when something said in a meeting is merely conversation and when it should trigger action. 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 remark as an instruction. It created two tasks assigned to QueueZero CEO Evan Wu: “Stop talking” and “Send pricing sheet.” Wu laughed, completed the second task and deleted the first. For a moment, the assistant had made its own chief executive an action item. QueueZero said Proxy requires confirmation before an external message is sent. Internal tasks can be created automatically, and administrators can disable automatic task creation. Those controls establish different thresholds for producing work inside an organization and communicating outside it. The distinction is significant as meeting assistants expand beyond transcription. A summary generally waits for a reader; an assigned task can enter a workflow immediately. When an assistant is also able to delegate follow-up work to software agents, decisions about permissions, review and automation become central to deployment rather than administrative afterthoughts. The demonstration concerned task creation, not the unauthorized sending of an external message. It nevertheless showed how an audience comment could be converted into an internal action. For customers, the relevant evaluation will therefore include not just the quality of Proxy’s notes, but how reliably it identifies intent and how easily its behavior can be constrained. Proxy’s design connects discussion more closely with execution. In that model, a meeting becomes a source of operational instructions: notes document the conversation, while tasks and agents carry out the resulting work. That can reduce the gap between agreeing on a next step and entering it into a system, but it also increases the cost of misunderstanding context. Proxy’s public beta will provide a broader test of that approach. The launch demo has already supplied a straightforward benchmark: an effective assistant must recognize action items without treating every remark as one.

Anthropic: Claude Fable 5.1: # QueueZero's New AI Meeting Assistant Assigns Its First Tasks—to the CEO QueueZero launched Proxy, an AI meeting assistant, on Wednesday with a live demonstration that doubled as an unplanned test of how the product handles ambiguous speech. Proxy is designed to join a meeting, generate notes, and assign follow-up work to software agents. Unlike earlier transcription tools, which stop at producing a summary, Proxy is meant to act on what it hears, turning spoken commitments into tasks that other automated systems can carry out. That capability was on display within minutes. During the demonstration, a customer in the audience remarked, “Someone should tell Evan to stop talking and send us the pricing sheet.” Proxy, which was listening to the room, interpreted the comment as an instruction. It created two tasks and assigned both to QueueZero chief executive Evan Wu: “Stop talking” and “Send pricing sheet.” Wu laughed, completed the second task and deleted the first. The demonstration continued. The exchange highlights a practical question facing products that act on conversational input: how a system decides which statements are directives and which are commentary. Meetings are full of offhand remarks, hypotheticals and sarcasm, and a tool that treats every imperative sentence as an assignment risks generating a backlog nobody asked for. In this case, one of the two tasks was arguably correct. QueueZero says Proxy includes safeguards against the more consequential version of this problem. According to the company, Proxy requires explicit confirmation from a user before any external message is sent, so it would not have emailed the pricing sheet on its own. Internal tasks, however, can be created automatically without a confirmation step. That is the setting that was active during Wednesday’s demonstration. Administrators can disable automatic task creation entirely, the company says, requiring a person to approve each item before it enters a team’s workflow. QueueZero did not say whether that setting will be on or off by default when customers first deploy the product. The distinction between internal and external actions reflects a pattern emerging across AI agent products, where vendors permit automation for low-stakes, reversible steps while placing human review in front of anything that leaves the organization. Whether creating tasks for colleagues counts as low-stakes is likely to depend on the organization. Proxy is priced at $30 per user per month. It enters public beta next week. For QueueZero, the launch arrives in a crowded category. Established collaboration suites already offer AI summaries, and several startups are pursuing agent-based follow-up. The company is positioning Proxy’s task-assignment features as its differentiator—a pitch that, intentionally or not, received a live demonstration on Wednesday.

MoonshotAI: Kimi K3.1 Pro: 6

Anthropic: Claude Fable 5.1: **6**

MoonshotAI: Kimi K3.1 Pro: # KiteBridge failure creates $2.8 million in duplicate reimbursements A payment-processing failure at KiteBridge generated 13,400 duplicate reimbursements worth $2.8 million on Tuesday, affecting 91 corporate customers and leaving a recovery effort unfinished the following day. Employers funded the erroneous payments, while no employees had money withdrawn from their accounts. KiteBridge makes expense-management software, which businesses use to process employee expenses and reimbursements. The incident therefore went beyond an administrative delay: its payment automation created duplicate transactions that had to be contained and then unwound. The sequence began at 8:45 a.m. Tuesday, when KiteBridge deployed version 4.2. The first customer complaint arrived at 9:12 a.m. The failure was traced to a timeout involving KiteBridge’s payment processor and the way the company’s fallback system responded. A fallback system provides a backup

Anthropic: Claude Fable 5.1: # KiteBridge Software Flaw Sends $2.8 Million in Duplicate Reimbursements to Employees at 91 Companies A software update from expense-management provider KiteBridge triggered 13,400 duplicate reimbursement payments on Tuesday morning, pushing roughly $2.8 million in unintended payouts to employees at 91 corporate customers before the company shut off its automated payment system. The money came from employers' accounts, not employees'. No worker had funds withdrawn, but many received the same expense reimbursement twice. By Wednesday evening, KiteBridge said it had clawed back $2.3 million, with the remaining half-million still pending recovery. ## A routine release goes wrong The trouble began at 8:45 a.m. Tuesday, when KiteBridge deployed version 4.2 of its platform. The release was not flagged as unusual, and the company has not said what the update was intended to change. Less than half an hour later, at 9:12 a.m., the first customer complaint arrived. Over the following hour, as reports accumulated, the company traced the problem to an interaction between its payment processor and its own fallback system. Expense software like KiteBridge's does not move money directly. It sends payment instructions to an outside processor, which executes the transfer. When that processor is slow to respond — a "timeout," in industry terms — well-designed systems typically wait and then try again, while making sure the retry is recognized as the same payment rather than a fresh one. That safeguard failed. According to KiteBridge, a processor timeout caused its fallback system to treat each retry as a new reimbursement request. Payments that had already gone through were sent again, and in some cases the original request and its retries were all honored. The result was thousands of employees receiving double payments for legitimate expenses. KiteBridge suspended automated payments at 10:03 a.m., about 78 minutes after the deployment and 51 minutes after the first complaint. ## A protection that was supposed to exist The failure is particularly awkward for KiteBridge because of its own marketing. An April changelog for version 4.0 announced the introduction of "exactly-once payment protection" — a feature meant to guarantee that a single reimbursement could never be paid more than once, regardless of network hiccups or retries. In software engineering, "exactly-once" delivery is a well-known and notoriously difficult promise, usually achieved by tagging each payment with a unique identifier so that duplicates can be detected and discarded. Whether version 4.2 broke that mechanism, bypassed it, or exposed a gap in it is not yet clear. KiteBridge has not explained how a system advertised as duplicate-proof produced 13,400 duplicates in just over an hour. ## Slow to disclose The company's communication during the incident has drawn separate criticism. KiteBridge's public incident page initially described the problem as *delayed* reimbursements — the opposite of what customers were experiencing. The page did not acknowledge duplicate payments until roughly 4 p.m. Tuesday, six hours after automated payments had been halted. For finance teams at affected companies, that gap mattered. Several were already fielding questions from employees and reconciling unexpected outflows while the vendor's official status page suggested payments were simply late. ## Recovery and next steps The recovery effort is still underway. Of the $2.8 million paid out in error, $2.3 million had been reversed or returned by Wednesday evening. KiteBridge has not detailed how the remaining funds will be recovered or whether employers will be made whole if some employees cannot or will not return the money. CEO Lena Ortiz apologized for the incident and said the company would publish a full technical report. She did not give a timeline for that report, nor for the restoration of automated payments, which remained suspended as of Wednesday evening. For now, the 91 affected companies are left reconciling their books — and weighing how much to trust a payment guarantee that, in practice, lasted less than four months.

MoonshotAI: Kimi K3.1 Pro: # Support AI Benchmark: Vale’s Value Is Clearer Than North’s Lead Model North earned the highest average score in a test of AI-generated customer-support answers, but the more useful finding concerns Model Vale. North scored 84.2 and Vale 83.8, a gap that was not statistically significant. Yet Vale’s estimated cost was $0.74 per 1,000 tickets, compared with $1.90 for North, and it answered 18% faster. For support leaders, that makes Vale the clearest value proposition: a near-top average in this test, at a much lower cost and with faster responses. That is not the same as declaring Vale universally best. The benchmark shows where averages help—and where they hide operational risk. ## How the models were tested Researchers gave five language models the same 200 anonymized customer-support tickets and collected one proposed answer per ticket from each model. Three experienced support managers graded the answers blindly. Their scores combined factual correctness, usefulness and tone, so the final averages reflect overall answer quality rather than any one dimension. Blind grading is a useful safeguard against being influenced by a model’s name or reputation. Still, one answer per ticket is a limited snapshot: it does not demonstrate consistency across repeated runs. A composite score can also conceal whether a model succeeded through accuracy, tone or some mixture of the three. ## The quality–cost trade-off Vale’s estimated cost was about 61% lower than North’s; equivalently, North cost roughly 2.6 times as much per 1,000 tickets. Because the score difference was not statistically significant, the benchmark provides little justification for treating North’s 0.4-point advantage as a meaningful quality premium. North may still be preferable in some settings, but that case needs more than the overall ranking. The wider field shows that cheaper answers can come with lower average scores. Ember paired a 79.1 score with a $0.31 estimated cost, while Kite scored 76.4 at $0.22. Harbor was cheapest at $0.08 but had the lowest score, 71.0. These results describe a trade-off in this sample, not a universal law that every price reduction costs a fixed amount of quality. Nor do the reported averages alone establish which of the other gaps are statistically meaningful. ## The right model may depend on the ticket North performed best on billing disputes, while Vale led on technical troubleshooting. That is a practical warning against choosing solely on the overall average. A support organization with a different ticket mix could reach a different conclusion, although the summary does not provide enough category-level detail to quantify that possibility. Category margins and significance tests are also needed before treating either specialty as a firm procurement rule. The strongest shared limitation is more clear-cut: all five models performed poorly when a ticket depended on an image attachment. A good average on the broader test therefore should not be read as evidence that any model can handle every support workflow. Image-dependent tickets need separate evaluation and, for now, an explicit path for human review. ## What the benchmark cannot settle The study used English-language tickets from one software company. Its results may not transfer to other industries, customer bases or writing conventions. Researchers also did not test agentic actions, tool use or multilingual support. This was an evaluation of drafted answers, not of end-to-end ticket resolution, and the estimated model costs should not be confused with the full cost of deployment, integration or human oversight. The practical takeaway is to put Vale on the shortlist for text-based answer drafting, while retaining North as a candidate where billing performance matters and treating the cheaper models as candidates for their own pilots rather than automatic substitutes. A follow-up on a company’s own tickets should measure category-level quality, repeatability, image handling and review effort. The benchmark’s central lesson is disciplined rather than flashy: the top numerical score did not establish a decisive winner. Vale’s lower cost and faster answers, combined with no statistically significant score deficit against North, did establish a strong value case.

Anthropic: Claude Fable 5.1: # The Cheaper Model Tied the Expensive One. That's the Whole Story. A new benchmark of five language models on real customer-support tickets produced a result that should get the attention of anyone paying for these systems: the second-place model matched the first-place model's quality within the margin of error, while costing less than half as much and answering faster. ## What the researchers did The team collected 200 anonymized support tickets from a single software company and had each of five models — North, Vale, Ember, Kite and Harbor — write one proposed reply per ticket. Three experienced support managers then graded every answer without knowing which model produced it. Their scores blended three things a real support lead cares about: whether the answer was factually right, whether it actually helped the customer, and whether the tone was appropriate. The researchers also tracked estimated cost per 1,000 tickets and response latency, which turns out to matter as much as the quality score itself. ## The headline: North and Vale are effectively tied North posted the top average score at 84.2. Vale came in at 83.8. That 0.4-point gap sounds like a ranking, but the researchers were explicit that the difference was **not statistically significant**. With 200 tickets and three graders, a spread that small is indistinguishable from noise — rerun the test with a different 200 tickets and the order could easily flip. What is *not* noise is the cost. North runs at an estimated $1.90 per 1,000 tickets; Vale runs at $0.74. That's roughly a 60% reduction for the same measured quality. Vale also returned answers 18% faster, which affects agent workflows and customer wait times in ways a quality score doesn't capture. Put plainly: on this benchmark, there is no evidence that paying North's premium buys better support answers. If a team is choosing between these two for general ticket handling, the data points to Vale. ## Where the two leaders differ The averages hide one genuine distinction. North did best on **billing disputes**; Vale did best on **technical troubleshooting**. The report doesn't give per-category confidence intervals, so these should be read as tendencies rather than settled facts, but they're plausible. Billing tickets reward careful, cautious wording about money and policy. Troubleshooting rewards procedural accuracy and willingness to walk through steps. A team whose queue is dominated by one category might reasonably weight that. A team with a mixed queue shouldn't over-read it: both models scored in the same band overall, and routing tickets by type to different models adds operational complexity that may not pay for itself. ## The rest of the field The remaining three models — Ember (79.1, $0.31), Kite (76.4, $0.22) and Harbor (71.0, $0.08) — trace a familiar curve where each step down in price costs a few points of quality. Ember is notable for landing within about five points of the leaders at less than half of Vale's cost, which could make it viable for high-volume, low-stakes tickets. Harbor is extraordinarily cheap but its 71.0 is far enough below the pack that the drop is unlikely to be measurement error. The gaps between these three are real but modest. Whether any of them is "good enough" depends entirely on what a wrong or awkward answer costs a given business. ## One weakness shared by all five Every model struggled when a ticket depended on an **image attachment** — a screenshot of an error, a photo of a broken part, a cropped invoice. None of the five handled these well. For many support operations, image-dependent tickets are a meaningful slice of volume, and this benchmark suggests none of the tested models is ready to own them without a human in the loop. ## What this benchmark doesn't tell you The limits are significant and the researchers listed them openly: - **One company, one language.** All 200 tickets were English and came from a single software vendor. Results may not transfer to other industries, product types or languages. - **Reply drafting only.** The models wrote answers. They didn't look up account data, issue refunds, call tools or take any agentic action — which is where much of the real work in support automation lives. - **Sample size.** Two hundred tickets is enough to separate a 71 from an 84. It's not enough to separate an 83.8 from an 84.2, as the researchers themselves note. - **Costs are estimates.** Pricing changes frequently and depends on volume and contract terms. ## Bottom line For drafting English support replies at one software company, two models reached the same quality tier, and one of them did it for 60% less money and 18% less time. Everything else in this benchmark — the mid-tier trade-offs, the per-category strengths — is secondary to that finding, and should be checked against your own ticket mix before you act on it.

MoonshotAI: Kimi K3.1 Pro: # Foundry Hall to Close, Leaving Twelve Startups Searching for Space Foundry Hall, a nonprofit startup accelerator in Milwaukee, will close October 31. Since opening in 2018, it has supported 212 startups, which collectively employ 1,140 people. Now, the organization that helped those companies develop is preparing to end its own work. Its financial problem centered on one relationship. Blank River Bank, Foundry Hall’s largest corporate sponsor, supplied 38 percent of its annual budget. When the bank declined to renew its sponsorship, the accelerator was left without enough money to continue beyond its current commitments. A bank spokesperson said the decision reflected a change in community investment priorities and was unrelated to Foundry Hall’s performance. The loss of that one funding stream proved decisive despite the accelerator’s record. Executive director Camille Ross said the organization had hoped other funding would arrive in time. “We kept waiting for the replacement check that never came,” she said. Foundry Hall has enough money to operate through October, but cannot afford another full program. Six employees will lose their jobs. Twelve startups currently using its workspace must leave, turning the closure into an immediate search for workspace as well as a longer-term loss of support. For founder Dev Malik, the building represents a specific turning point. “My first investor meeting happened in that building,” he said. “Before Foundry Hall, I was pitching from the back table of a coffee shop.” His account puts the accelerator’s work on a smaller scale than its cumulative numbers. The 212 startups and 1,140 jobs show its reach. They do not fully convey the early, uncertain stage when a founder needs a place to work, practice a pitch or meet someone willing to listen. For Malik, Foundry Hall was part of that passage from improvisation to legitimacy. City officials are concentrating on the companies still in the building. Deputy mayor Susan Cole said the city is speaking with building owners and local universities about space for them. Those conversations could provide the twelve startups with their next locations, but they will not save the accelerator’s program or preserve its six positions. That leaves Foundry Hall with an ending measured in practical tasks: winding down operations and emptying workspace that twelve companies still use. Its record will remain in the businesses it assisted and the people they employ. But the closure also removes a setting for work that is harder to count — the early, uncertain stage when a founder needs a place to be taken seriously before the company’s future is secure. Malik’s memory makes that loss tangible. For him, Foundry Hall was the distance between a coffee shop’s back table and a room where his first investor meeting could happen.

Anthropic: Claude Fable 5.1: # Foundry Hall to Close Oct. 31 After Largest Sponsor Walks Away On a weekday morning at Foundry Hall, the whiteboards are still full. Someone has sketched a pricing model in blue marker; someone else has left a half-erased org chart. In a few weeks, the boards will be wiped for good. The nonprofit startup accelerator in Milwaukee will close October 31, ending a run that began in 2018 and helped launch 212 companies now employing a combined 1,140 people. The cause is a single missing line item. Blank River Bank, the accelerator's largest corporate sponsor, supplied 38 percent of its annual budget and declined to renew its sponsorship this year. Foundry Hall has enough cash to keep the lights on through October. It does not have enough to run another full program, and without a program there is little reason to keep the doors open. "We kept waiting for the replacement check that never came," said executive director Camille Ross. A sponsor that supplies more than a third of a budget is not easily replaced on short notice, and Ross made clear the organization did not find a substitute in time. The money that remains is enough for an orderly wind-down, not a rescue. A bank spokesperson said in a statement that the decision reflected "a change in our community investment priorities" and was "unrelated to Foundry Hall's performance." The statement did not say what the new priorities are. Six staff members will lose their jobs. Twelve startups currently working out of the building must find new space before the end of October. For some of them, the desks at Foundry Hall were the only office they had ever had. Dev Malik, a founder who came through the program, said the building mattered in ways that are hard to capture on a grant report. "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 shift — from coffee shop to conference room, from a laptop balanced on a knee to a table where someone was willing to sit down across from you — is what the accelerator sold, and what the numbers only partly describe. Over 1,100 jobs is a concrete figure. The confidence it took to create them is not. City officials say they are trying to soften the landing for the displaced companies. Deputy Mayor Susan Cole said the city is in conversations with building owners and local universities about workspace for the twelve startups. She did not offer a timeline or say whether any arrangement would be subsidized. Nobody at Foundry Hall is pretending that a new desk somewhere else will replicate what the accelerator did. Mentoring, demo days, introductions to investors — those depended on six people who will be gone by November. For now, Ross is focused on helping the current companies move out, and on closing the organization cleanly rather than letting it fade. The whiteboards, for the moment, stay as they are.

MoonshotAI: Kimi K3.1 Pro: # Raincheck Raises $8 Million in Seed Funding *The Austin company’s flood-risk software helps local governments decide where to send maintenance crews before storms.* Raincheck, an Austin company that makes flood-risk software for local governments, has raised $8 million in a seed round announced today. Basin Ventures led the round, with participation from existing investor Long Mile. The new funding brings Raincheck’s total raised to $10.5 million. Raincheck maps drainage systems and combines those maps with rainfall forecasts. Cities use the software to decide where to send maintenance crews before a storm. The company provides additional detail about its approach in its [methodology documentation](https://example.com/raincheck-methodology). Raincheck CEO 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. Raincheck currently works with seven cities in Texas and Louisiana. The company declined to identify them because some contracts have not been publicly announced. The company’s forecasts are recommendations, and emergency managers retain control over deployments and public warnings. ## Competition and Expansion 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 Prepare for Floods Before the Rain Falls *The Austin startup's software tells public works departments which drains, pumps and repairs to prioritize ahead of a storm.* Raincheck, an Austin company that makes flood-risk software for local governments, said today it has raised $8 million in seed funding. The company 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. Basin Ventures led the round, with participation from existing investor Long Mile. Raincheck has now raised $10.5 million in total. CEO Ana Flores said the challenge for most cities is not identifying flood-prone areas but ranking the work that needs to happen before a storm: > "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 currently works with seven cities in Texas and Louisiana. It declined to identify them, saying some contracts have not yet been publicly announced. Its forecasts are recommendations rather than directives. 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.

MoonshotAI: Kimi K3.1 Pro: # HelioDock launches Night Shift robot monitoring and raises $24 million *The remote fleet-monitoring software will cost $90 per robot per month when it becomes generally available September 15, after tests involving nine logistics companies and 310 robots.* HelioDock on Thursday launched Night Shift, software that lets operators monitor fleets of autonomous warehouse robots remotely. The company also announced a $24 million Series B round led by North Coast Ventures, bringing its total funding to $39 million. The launch moves Night Shift from private testing toward a paid rollout, with HelioDock citing results from the beta as evidence of an operational benefit. The company says customers experienced fewer unplanned stoppages, but one beta user also described a clear constraint: the system worked well where connectivity was strong and became less useful during rural LTE outages. Night Shift therefore arrives with both a performance claim and a practical picture of when remote oversight is most dependable. CEO Mara Velez described the problem as an unwanted trip triggered by a single machine. “A warehouse manager should not have to drive across town at two in the morning because one robot

Anthropic: Claude Fable 5.1: # HelioDock Launches Remote Monitoring Software for Warehouse Robot Fleets *Night Shift, tested for 14 months across nine logistics companies, arrives with a $24 million Series B led by North Coast Ventures* HelioDock, a maker of autonomous warehouse robots, on Thursday released Night Shift, software that lets operators monitor and intervene in robot fleets without being on site. The company paired the launch with a $24 million Series B led by North Coast Ventures, bringing its total funding to $39 million. The product targets a routine but costly problem in automated warehouses: a robot encounters something it cannot interpret, stops, and stays stopped until a person shows up to resolve it. Night Shift moves that intervention off the warehouse floor and onto a remote operator's screen. HelioDock is betting that logistics companies will pay a monthly per-robot fee to avoid late-night site visits and the idle time that accumulates while a stalled machine waits for help. "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. Night Shift will be generally available September 15 at $90 per robot per month. For a fleet the size of the beta program, 310 robots, that works out to roughly $27,900 a month, or about $335,000 a year. HelioDock developed the software over a 14-month private beta with nine logistics companies operating those 310 robots combined. The company says participating customers experienced 17% fewer unplanned robot stoppages during the beta. That figure is the company's central argument for the price, and it has not been independently verified. Nine companies is also a modest sample, and the notes HelioDock has shared do not indicate how results varied from one facility to the next. Feedback from the beta points to connectivity as the deciding factor in how much the software delivers. Aaron Pike of Fulcrum Logistics, one of the participating companies, said Night Shift worked well in facilities with strong connectivity but became less useful during rural LTE outages. The reason is structural. Night Shift depends on a live link between the robot and the remote operator. When that link drops, the robots revert to whatever behavior has been configured locally on each machine, and remote control is unavailable until the connection returns. A disconnected robot is not left without instructions, but the person watching from across town loses the ability to do anything about it. In a facility with spotty cellular coverage, the software's absence can coincide with exactly the kind of overnight stoppage it is designed to address. For operators, the practical question is less whether the product works than where. Warehouses with reliable connectivity stand to get the most from it. Sites that lean on intermittent rural LTE will get less, as Pike's experience suggests, and will need to be comfortable with the fallback behavior their robots are configured to follow when the link goes down. Night Shift also enters a market that already has options. Relay Robotics and YardOS each sell competing fleet-management products, and HelioDock will be asking operators to adopt its software at $90 per robot per month against those alternatives. The 17% stoppage reduction, if it holds outside the beta, is the number HelioDock will have to defend in those conversations. The Series B gives the company more capital to make that case. North Coast Ventures led the round, which more than doubles HelioDock's cumulative funding from the $15 million it had raised before Thursday. The company has framed the raise and the product launch as a single announcement, tying its new funding directly to Night Shift's move from a controlled beta to general availability. Velez's pallet scenario captures the pitch in one sentence. Whether it captures the experience of customers whose warehouses sit beyond reliable cell coverage is a question the first wave of paying users will answer after September 15.

MoonshotAI: Kimi K3.1 Pro: # 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.

MoonshotAI: Kimi K3.1 Pro: # Cloudnote Debugging Page Exposed Customer Prompt Excerpts for Four Hours *The company says it has found no evidence of access beyond the researcher who reported the issue, but its investigation is not complete.* Cloudnote discovered that a debugging page containing customer prompt excerpts was accessible without authentication for four hours on Monday. Logs show excerpts from 1,842 user sessions, and the excerpts could include up to 200 characters from a prompt. The company has 3.2 million registered users. Cloudnote removed the page 19 minutes after a researcher notified it. The company says it has found no evidence that anyone besides the researcher accessed the page. That finding is not proof that no other access occurred; Cloudnote has not completed its investigation. Full documents, passwords and payment information were stored elsewhere and were not present on the page. Even so, short prompt excerpts can contain sensitive details depending on what users entered, so the character limit alone does not establish the severity of the exposure. What those excerpts contained remains unclear. The session count also does not establish how many individual users were affected. Key questions remain about how the page became accessible and whether anyone beyond the researcher viewed or copied the material. Those answers will determine the incident’s practical impact for the users whose sessions appeared on the page.

Anthropic: Claude Fable 5.1: # Cloudnote Exposed Excerpts of Customer Prompts on Unprotected Debugging Page *The company says the page sat open for four hours Monday and showed snippets from 1,842 user sessions. It has found no sign that anyone but the researcher who flagged the problem saw it — but its investigation is not finished.* Cloudnote left an internal debugging page accessible to anyone on the internet, without a login, for about four hours on Monday, exposing short excerpts of prompts that customers had submitted to the service, the company said. According to Cloudnote's logs, the page displayed snippets from 1,842 user sessions. Each excerpt could contain up to 200 characters of a prompt — roughly a sentence or two. Full documents, account passwords and payment information are stored on separate systems and did not appear on the page, the company said. A security researcher discovered the page and notified Cloudnote, which removed it 19 minutes later. The company said it has found no evidence that anyone other than the researcher accessed the page while it was exposed. It has not completed its investigation, however, and that finding could change as the review continues. The affected sessions represent a small fraction of Cloudnote's 3.2 million registered users. Still, prompts are often where people state plainly what they are working on — a client's name, a medical question, a draft of a difficult message — and even a 200-character fragment can carry sensitive detail. Cloudnote has not said what the exposed excerpts contained, how the page came to be publicly reachable, or whether it plans to notify the users whose sessions appeared.

MoonshotAI: Kimi K3.1 Pro: C

Anthropic: Claude Fable 5.1: C