Somebody Still Has to Grade the Answer

Blue J

Blue J is designing its next AI product to stop and ask the tax professional a question. In a business paid to exercise judgment, that pause deserves a place in the product demo.

The Toronto company announced a $176M Series E on November 10, led by JMI Equity, with Oak HC/FT, Sapphire Ventures and Intrepid Growth Partners participating. Blue J puts its valuation above $700M. JMI Principal Jack Duane will join the board as the company builds beyond finding tax answers into preparing the work clients actually need.

CEO Benjamin Alarie and CTO Brett Janssen co-founded Blue J in 2015 with Anthony Niblett and Albert Yoon. That combination of tax-law expertise and software engineering now has another assignment: help a practitioner carry an answer through the documents, assumptions and decisions that turn it into advice.

The forthcoming Workflows product is designed for return reviews, IRS notice responses, partnership and trust agreements, multistate nexus analysis and client communications. Blue J says the steps and sources will be visible, with requests for professional input when a judgment call arises. Workflows remains described as soon-to-be-released, with a waitlist.

A partner reviewing a draft needs to see how it got there. Otherwise, faster preparation can simply deliver a larger pile of material to the most expensive person in the process. Inspectable reasoning gives that reviewer somewhere to enter the work, challenge an assumption and keep moving toward the client's answer.

Blue J reports 165% ARR growth in the 15 months since its $122M Series D, more than 6,500 organizations using the platform and over 14M tax questions answered. Those figures describe the research business that earned this financing. The next product has to earn its place in the firm's delivery process.

The round also includes secondary share purchases, according to BetaKit's interview with Benjamin Alarie. The split was undisclosed, so the full $176M should not be treated as fresh operating cash. Blue J plans to scale Workflows and expand in the UK and internationally.

Its October partnership with Legora gives eligible joint customers another way into specialist tax research, inside the software where their legal matters already live. Distribution and deeper preparation are advancing together. For the partner who eventually signs off, the useful promise is being able to spend less of the review reconstructing how somebody reached the recommendation.

Alterra Labs

Attending physicians are reviewing the grading rubrics at Alterra Labs. That is a fairly consequential use of a doctor's time: deciding what an AI model should be rewarded for before it gets another chance to do the work.

Flare Capital Partners describes multiple physicians reviewing each rubric, with clinicians and machine learning researchers examining where a model's sequence of actions went wrong. The product includes the judgment about the attempt, right down to the parts a plausible final answer might hide.

Founder and CEO Akash Kapoor has raised a $12.5M seed round for that work, announced November 10. Oak HC/FT led, with Flare Capital Partners, Company Ventures, Tau Ventures and angels participating. The New York company builds simulation environments, evaluations and post-training infrastructure for healthcare and life sciences.

Akash Kapoor came through machine learning research at the NIH and medical training at Columbia. Dr. Morgan Jeffries, Chief Clinical Officer, brings experience as Geisinger's former Medical Director for AI and leads Alterra's clinical fellowship program. Together, those backgrounds put people who understand the work inside the process of deciding how software should learn it.

The environments span electronic health records, imaging, payer workflows, clinical trials and scientific research. A model has to navigate the task, while the people designing the evaluation decide what counts as sound reasoning and an appropriate action. Alterra's own standard rejects a correct result reached for the wrong reasons.

That gives the business an interesting pair of customers. A frontier lab wants feedback that helps improve its model. A health system wants evidence that helps decide which model to use, under which safeguards. Both need an evaluation close enough to their actual work to deserve attention.

Flare reports paid partnerships with a handful of frontier labs, without naming them or disclosing contract values. Alterra says the funding will expand research, its in-house clinical and scientific team, and partnerships. Those priorities keep specialist knowledge close to the machinery producing the training signal. Healthcare expertise becomes something the model builder needs again as the work changes.

For a hospital choosing AI, the useful output includes an explanation of where the model struggled. For the researcher receiving that explanation, the next job is to build a better attempt around the clinical judgment that exposed the failure.

Bay

Someone has to answer when an AI agent asks permission. Bay is building for the security team on the receiving end, where a sensible-looking request can carry the accumulated risk of everything the agent did before it.

Bay emerged from stealth on November 10 with a $37M Series A led by Battery Ventures, alongside a previously undisclosed $13M seed led by YL Ventures and Striker Venture Partners. That brings disclosed funding to $50M. Micron Ventures, M12, Microsoft's Venture Fund, and Wiz CEO Assaf Rappaport joined the Series A.

CEO Yuval Zacharia founded the Tel Aviv company with CPO Aner Morag and VP of R&D Boris Danilovich. Their backgrounds span Hunters, Noname Security and CrowdStrike. The company says its own coding agent once tried to expose a cloud environment to the internet while completing a test. It had permission to do the work. The proposed route created the danger.

That puts a demanding question inside the product: how much of the work must security understand before deciding whether the next action belongs there? A command viewed alone leaves out the task, the earlier steps and the resources the agent can reach.

Bay says it follows that session and evaluates proposed actions locally, with choices to allow, ask, deny or steer. Discovery maps agents and their tools; runtime enforcement carries policy into the moment those tools are used. The buyer is paying for a way to let work continue while keeping consequential decisions within the organization's rules.

The interesting detail is the rehearsal. Bay's Lighthouse system draws on 30 days of agent history to help build policies, and teams can simulate their effects before enforcement. The policy has a human workload attached to it. That gives security staff a way to examine which requests would interrupt people before those interruptions become part of everyone's working day.

Bay names Micron, Peloton, ClickUp and Wix among its customers. Micron's investment sits beside that customer relationship. The company's deployment and response-speed claims still need evaluation in the buyer's own environment, with its particular tools, data and tolerance for interruption.

The capital will expand product and engineering, plus sales, marketing and customer teams in the US and Europe. As those customers give agents longer assignments, Bay's policy decisions will help determine which work gets finished autonomously and which request lands with a person who has to understand it.

Mercor

Mercor wants an enterprise's own standards to travel with the AI agent doing its work. A model can know a profession and still miss the way a particular business expects a job to get done.

Brendan Foody and Adarsh Hiremath, Mercor's co-founders and co-CEOs, are taking that problem beyond the frontier labs. Mercor announced a $750M Series D at a $20B valuation on November 10. The transaction includes an employee tender; the company has not disclosed the split between new capital and employee liquidity.

Benchmark, Felicis, Fundomo, funds managed by affiliates of Apollo, G Squared, General Catalyst, Glade Brook Capital, NVIDIA, Salesforce Ventures, Sapphire Ventures, StepStone Group, XTX Ventures and Zoom participated. Mercor's October 2025 Series C was $350M at a $10B valuation, led by Felicis. The new announcement names participants without designating a lead.

Founded in 2023 by Brendan Foody, Adarsh Hiremath and Surya Midha, Mercor started in hiring. Its ability to find and assess specialized workers became useful to companies training AI. Now the buyer can be the enterprise that needs an agent to understand its systems, policies and quality bar.

That changes the conversation with the expert. Knowing how to do the work helps establish what a useful evaluation should demand. Knowing the company helps explain which apparently reasonable answer could still create a problem for the person responsible for the result. That translation carries the buyer's operating judgment into the test.

Mercor acquired Deeptune this year to add simulated enterprise applications where agents can practice. Expert-designed tasks, a place to attempt them and a way to judge the outcome can sit inside the same business. The acquisition gives Mercor more of that training loop to build and sell.

The company reports more than 5M experts in its network, a measure of reach rather than the number working on any customer's project. Its next investment goes into evaluation and post-training infrastructure, research and more environments for complex workflows. Offices in New York and London extend the operation beyond San Francisco.

For an enterprise buyer, the work starts with the standards it can explain to Mercor's team. A finance policy, an escalation decision or a quality threshold has to become something an agent can practice and a reviewer can assess. As those assignments grow longer, the people who understand the business will keep deciding what counts as a finished job.

Nanopath

More than 100 healthcare providers helped Nanopath's founders decide where their diagnostic technology belonged. Women's health emerged from those conversations as a place where a faster answer could change the shape of the appointment.

That history, described by Norwest in 2022, gives the company's latest financing some depth. Co-founders and co-CEOs Alison Burklund and Amogha Tadimety are now advancing a 15-minute point-of-care vaginitis assay toward clinical study and potential commercialization.

Nanopath announced $25M in Series B financing on November 10, including $2M in non-dilutive grants. Norwest and SV Health Investors' MedTech Convergence Fund co-led, with Labcorp Venture Fund, Elderberry Ventures, GingerBread Capital and Alumni Ventures participating.

The product's time target carries a practical ambition. A patient and clinician could have molecular test information available while they are still together, with the conversation and the result inside the same visit. Nanopath is designing around that encounter.

Its optical biosensing approach detects molecular targets without the nucleic acid amplification used in PCR. The company describes a cartridge-and-reader system intended to bring laboratory-quality testing to the point of care. The vaginitis assay remains in development and requires applicable regulatory review before commercial availability.

This raise funds work that has to travel together. Manufacturing scale-up needs a repeatable product. The planned FDA 510(k) clinical study needs evidence. Clinical and commercial hiring brings more people into preparing the company for a potential launch.

Burklund and Tadimety both earned engineering PhDs at Dartmouth. Their decision to spend time with providers put a clinical setting around the science early, giving the team a specific use case to work toward through years of development.

Norwest and the MedTech Convergence Fund also co-led Nanopath's $10M Series A in 2022. Their return connects the earlier technology thesis with a more operational stretch of work, where the company has to prepare its test and organization for use beyond the development team.

For a clinician considering a point-of-care test, the useful unit of progress is an appointment in which the result arrives soon enough to inform the conversation. Nanopath's new financing follows that ambition into manufacturing, clinical evidence and the people preparing a potential launch.

Charta Health

Charta Health makes its growth team learn medical coding. That puts a practical limit on what the sales pitch can get away with.

CEO Justin Liu and CTO Scott Morris arrived from Rockset with data infrastructure experience. Before building Charta, the co-founders spent a year earning Certified Professional Coder credentials and interviewing more than 100 healthcare leaders. Engineering and growth employees now have to get certified, too.

There is a commercial choice inside that homework. Someone selling chart review has to understand why a diagnosis belongs on a claim, why another one should come off, and what the documentation can actually defend. The customer needs an explanation that survives a coding review after the demo ends.

Insight Partners is backing that work with a $55M Series B announced November 10. Madrona, Bain Capital Ventures and South Park Commons returned, with QuantumLight also participating. Charta reports $85M raised across 3 rounds in under 2 years and revenue growth of 10x over the last 12 months.

The San Francisco company reviews medical charts before billing, checking documentation against coding guidelines, payer rules and clinical protocols. Straightforward fixes can be automated; more nuanced issues reach people inside the electronic health record. A clinician or coding team can deal with the exception while the record is still close to the encounter.

Charta already names American Family Care, Grow Therapy and Pediatric Associates Family of Companies among its customers. Its next expansion puts more weight on the other institution reading the chart: the health plan.

Charta for Providers and Charta for Payers run on the same AI engine. The payer offering begins with Medicare Advantage and ACA Marketplace plans, covering risk adjustment, coding checks, audit preparation and quality gaps. It was developed with a large insurer whose identity remains undisclosed.

That arrangement asks the company to earn confidence from buyers with different financial interests. The provider wants payment for documented care. The health plan needs support for the diagnoses it accepts. Charta's proposed shared standard has to remain useful when a supported correction changes which side likes the number.

The funding supports AI development, more specialties and that payer expansion. As Charta's engineers and growth team take the product into another organization, their coding education travels with them into the conversation about what the chart can support.

Nidus

The factory's knowledge has an owner. Nidus is building its robotics business around keeping that knowledge with the defense manufacturer, even as AI learns more of the work happening inside the plant.

That choice matters commercially. A customer buying more production capacity also has to decide who gets to learn from its processes. Nidus says its sovereign training practices let each manufacturer retain its collective knowledge. In this market, the data relationship belongs in the sales conversation alongside the robot.

CEO Justin Sealy, CTO Charu Dwivedi, president Harrison Nguyen and fellow co-founder Cane Punma have raised $30M to bring embodied AI into America's defense factories. Announced November 10, the total spans a Costanoa-led pre-seed and a Kleiner Perkins-led seed, according to Upstarts Media. The split between those financings remains undisclosed.

The team starts with wire harnesses, the bundles of wires, cables and connectors that travel into aircraft, missiles and vehicles. Nidus develops models that let industrial robots handle precision assembly, with wire processing as its first deployment focus. The machine has to learn work that already consumes somebody's time.

Kleiner Perkins partner Aatish Nayak is backing a team whose access to demanding customers is part of the investment logic. Costanoa founder Greg Sands points to increasing output inside existing factories. Those are useful perspectives on where the value might sit: close enough to production to understand what the manufacturer needs done.

Nidus says revenue passed 7 figures in its first 4 months. Upstarts describes more than $1M in bookings over that period. Those measures should stay separate; neither source provides audited financials or a named customer roster. The disclosed traction gives the financing a commercial context without supplying a complete account of the business.

The funding will expand engineering, support additional factory workflows and scale US defense deployments. Nidus is hiring in person in New York, putting model development and the practical work of deployment inside the same growing company.

For the manufacturer, protecting knowledge and increasing output now arrive in the same purchase. As Nidus moves beyond wire harnesses, each additional assembly process brings another conversation about what the customer will share, what its people need to teach and how much of their day the robot can give back.

Path Robotics

Andy Lonsberry wants Path Robotics to follow the work beyond the welding station. A factory that can weld more still has to assemble, inspect and finish what comes through.

That is where the Columbus company's $300M Series E starts to get interesting. Announced November 10 at a $2.45B post-money valuation, the Georgian-led round will help scale welding deployments and fund the data, compute and model development for more manufacturing tasks and robot types.

CEO and co-founder Andy Lonsberry and CTO and co-founder Alex Lonsberry grew up welding with their father. Their company built Obsidian to handle the part variation that makes a factory floor such an inconvenient place for a rigid script. The system finds seams, plans welds and adjusts to the metal in front of it. Every imperfect fit is part of the job the customer needs done.

Path's intelligent welding cells put that capability inside production. Rove brings it to the workpiece on a legged platform, useful when the structure is too large or awkward to bring into a cell. Saronic is an early adopter in its Franklin, Louisiana shipbuilding operations. Path's Rove page targets 50 units shipping in 2027, leaving deployment work ahead of the announcement.

Path also sells its welding cells as a service, with no upfront equipment purchase and 24/7 support. That keeps the supplier involved after installation, when a new part or a different shift has to make the machine productive again.

HII gives the expansion a demanding customer context. Its HYPR agreements contemplate up to $900M of work across Path Robotics and GrayMatter Robotics over 7 years, contingent on readiness and performance. Path describes its portion as $600M. Development and qualification come before delivery, with cost, schedule and quality governing the work that follows.

Georgian joins Addition, Drive Capital, IronArc Ventures, Tiger Global, Yamaha Motor Ventures and affiliates, D. E. Shaw Ventures and NAVER Ventures in the round. Georgian partner Emily Walsh will join the board. Path reports 17x ARR growth in 18 months, a useful adoption signal whose starting revenue base remains undisclosed.

The commercial opportunity keeps widening around the weld. A shipbuilder buying more capacity needs the next operations ready to absorb it, and the AI supplier has a reason to keep following that demand through the factory. The same customer that made welding worth automating can bring assembly and inspection into the next conversation.

Clara

Finance teams are being offered AI that can spend money. Clara is putting the approval rules in the same conversation, which gives its new funding a more interesting job than buying another dashboard.

Gerry Giacomán Colyer, cofounder and CEO, and Diego Iván García Escobedo, cofounder and CTO, built the Latin American payments company around businesses that need spending visibility and a practical way to act on it. Clara now brings corporate cards, bill payments, expenses, and accounts payable and receivable into one platform.

The November 10 financing separates two kinds of ambition: a $65M Series C anchored by Citius, Kaszek and DEG, plus a renewed $40M growth facility from General Catalyst’s Customer Value Fund. The combined $105M announcement includes that renewal. Calling the whole package a Series C would give the equity round an extra $40M it never had.

Clara says the capital will expand its go-to-market team and strengthen automation and AI. Its CEO also describes hiring local sales representatives across Latin America. That is a useful detail from a company selling software that has to understand how a customer’s finance department actually operates.

A finance team can ask for a spending report in natural language. Through Clara’s Model Context Protocol connections, an AI agent can also execute an authorized transaction. Policy checks review expenses and send exceptions to people. The distance between answering a question and moving company money is where permissions start earning their keep.

Clara reports more than 45,000 customers, over $100M in annualized net revenue and more than a year of positive EBITDA. Those are company-reported figures; annualized revenue describes a run rate, and EBITDA leaves out expenses that still matter. The traction gives this expansion a substantial operating base without turning the announcement into an audited financial statement.

Returning investors include Acrew Capital, Citi Ventures, Coatue, Commerce Ventures, DST Global Partners, Gaingels, General Catalyst, Mana Ventures, Monashees and Notable Capital. New participants include CMT Digital, Nascent VC and Michael Gilroy. They are backing a platform whose next sale may involve explaining exactly which decisions a customer can safely delegate.

For Clara’s local sales teams, the AI conversation can get very specific: which payments move automatically, which exceptions reach a person, and who remains accountable when the month closes.

Zus Health

The patient has become healthcare's most portable filing system. Ask someone to remember a medication change, a specialist visit and a test from another hospital, and a data-integration problem quietly becomes their homework.

Jonathan Bush built Zus Health around getting that history to travel with the person. The Boston company announced a $33M Series B on November 10, led by Drive Capital, bringing reported total funding above $100M. Sanofi Ventures, HealthX Ventures and Alumni Ventures joined returning investors a16z, F-Prime, Maverick Ventures and JAZZ Venture Partners.

There is an unusually practical business underneath the AI language. A healthcare application needs records from outside its own walls, then someone has to keep those connections working and the information current. A new summary feature can inherit an old maintenance department before it ever reaches a clinician.

Zus packages that work into a national data network, the Zus Aggregated Profile and clinical workflows. EHR records, pharmacy information, labs and care events contribute to a longitudinal patient record. Generated Patient Summaries sit on that foundation, so the software explaining the history has more of the history available to explain.

Jonathan Bush, the athenahealth co-founder, leads Zus as CEO; Philip Wickline is CTO. Their architectural choice gives customers a shared starting point for applications rather than another collection of connections to maintain separately. The useful question for a buyer is how much of that work actually disappears from the team's day.

Zus reports more than 110M networked transactions in August and 39M evidence citations year-to-date. Those are different activity measures, supplied by the company. Keep them in their own lanes: neither number tells a provider how much time a particular deployment will save.

Zus also announced QHIN candidacy in August, a step toward TEFCA exchange. The financing is earmarked for clinical intelligence, broader and fresher network coverage, and more workflows. That puts engineering effort behind the information arriving before a visit, surviving a care transition and reaching the application somebody already uses.

For a digital health builder, the purchase reaches beyond a summary on a screen. It includes the continuing job of finding the next record, reconciling it with the last one and making it available when the patient sees someone new. That work follows the patient long after the demo ends.

OpenAI / Odessia

Street noise made it into Odessia's hotel research. That tells you something about the travel business Francis Davidson is taking into OpenAI.

OpenAI has acquired the AI travel startup, and Francis Davidson announced on November 10 that he is joining the company to work on travel. Odessia previously announced a $6M round led by Sequoia Capital. The acquisition price has not been disclosed.

The product reached past a handsome itinerary. Odessia combined flights, hotels and experiences in a conversation, with preferences carried into the search and options people could actually book. Sequoia's description put transportation, places to stay and price comparisons inside the same experience.

In his October launch, Francis Davidson described hotel research that reached into breakfast menus, gym equipment and the noise outside. Those details sound small until you are the person spending a week with them. A room can satisfy the search filters and still be a lousy match for the trip.

That gives a specialist travel agent a demanding job beside the model. It has to turn a request into choices that survive contact with availability, price and the traveler's actual preferences. Someone may accept a longer flight to keep a hotel they love, or trade a preferred neighborhood for a better room. The agent needs to carry those decisions through the booking.

Francis Davidson had already co-founded Sonder. With Odessia, he returned to travel through software that helped people choose and book, bringing another view of the distance between a recommendation and a stay someone would happily pay for.

For a model company, that experience is worth understanding. Travel asks software to remember what a person values while dealing with options that change. The person who built the product has had to make decisions about that whole experience, including where the software needs more information before it acts.

OpenAI's plans for Odessia's technology remain unannounced. The immediate customer consequence is visible: Odessia's standalone service is no longer available, and its site directs existing bookings and travel plans to support. An acquisition can move the work forward while leaving users with a very practical question about the trip already on their calendar.

Francis Davidson is joining OpenAI with the experience of putting those details into a product. A traveler trying to sleep beside a noisy street can supply a fairly exact review of how well an agent understood the assignment.

Lumbra

Knowing the answer and being allowed to share it are different jobs inside an intelligence agency. Lumbra is building for the analyst who has to do both, across systems that each have their own rules.

CEO Aaron Brown, a former CIA officer, and CTO Michael Hiebert, an MIT-trained engineer, founded Lumbra in 2025 after careers that included service as Army Rangers. Their starting point put classified data and the person responsible for using it inside the product requirements.

Lumbra announced a $25M Series A on November 10, co-led by Point72 Ventures and Ansa Capital, with Costanoa Ventures returning. Most of the proceeds will fund engineers and deployment teams taking its agent platform into regulated commercial sectors. For the buyer, deployment expertise becomes part of the product being purchased, alongside the software.

An AI vendor can sell a model's capability in a demonstration. Putting agents inside an institution means learning which sources they can access, which conclusions somebody will accept and who gets to approve a change to the official record. There is a lot of customer knowledge hiding in those decisions. An institution's access policy has to survive that handoff.

LumbraOS divides that work among 3 products. Nebula coordinates agents and traces claims to sources. StarChart keeps findings in a knowledge graph with sources and sensitivity labels. Astrolabe evaluates agents before deployment and during production against custom benchmarks.

The company says domain experts validate results before agents write to a system of record. Customers own their knowledge graphs and can export them. The specialist's judgment becomes part of the workflow, while the institution retains the work it has helped establish.

Ansa says Lumbra won its first intelligence-agency contract 2 months after product work began, then reached production on a later government contract in 6 months. A June government announcement separately named Lumbra in the Pentagon's Agent Network project. Those disclosures establish a meaningful starting market; commercial expansion still requires its own customer work.

Finance, healthcare and law are among the intended destinations. Each brings a different vocabulary for acceptable evidence, different access rules and a different person who has to live with the decision. Lumbra's next deployment teams will be translating that vocabulary into what an agent can read, retain and put in front of a practitioner for approval.