The Money Is Chasing the Work

Siena AI

Brands have spent years teaching support teams to resolve the conversation in front of them. Siena AI is betting that the larger value begins after resolution, when what the customer said can change the next sale, policy, campaign, and product decision.

Siena raised a $17M Series A led by York IE, with Joyance Partners, AGLAÉ VENTURES, SIERRA Ventures, and SuperAngel.Fund participating. The round brings company-reported total funding to $29.7M.

Andrei Negrau, cofounder and CEO, and Lisa Popovici, cofounder and President, started with an empathic customer-service agent. It can handle a late delivery, address change, return, or subscription request while preserving policy, tone, context, and a path back to a person.

Hundreds of deployments exposed the larger problem: resolving the interaction does not guarantee the company keeps what it learned.

A shopper may explain that a sweater was right but the fit was wrong. That detail should improve the next recommendation, surface a pattern for the product team, and change how another agent handles a similar customer. Instead, it often disappears into the transcript that closed with the ticket.

Siena calls its next phase the Agent of Record for consumer brands. Siena Intelligence combines conversations with orders, subscriptions, reviews, loyalty data, and brand knowledge. Siena Memory carries that context into Shopping, Social, and voice agents, while human teams use a shared inbox, quality evaluation, testing, and Autopilot.

This is an organizational argument disguised as a product roadmap. Support, shopping, social, retention, and marketing may report to different leaders and live in different tools. The customer experiences one brand.

Siena says it works with hundreds of brands, including FIGS, SPANX, HexClad, and Comfrt, and has automated hundreds of millions of interactions. Its pages report up to 80% automation and a 4.81 average CSAT out of 5. Those figures are company-reported, but they show the production environment behind the expansion.

The capital will fund more autonomous agents and the infrastructure underneath them. A useful Agent of Record has to remember enough to improve the relationship, respect policies across channels, leave its reasoning open to inspection, and give people control when the interaction reaches its boundary.

Customer experience software optimized the handoff between a person and a ticket. Siena is financing the handoff between one customer moment and the next decision the brand makes.

Ampersand

The logo says Salesforce. The implementation says 12 years of custom objects, renamed fields, permission rules, and one workflow nobody is willing to touch.

That gap is where Ampersand is placing a $15M Series A led by Bessemer Venture Partners.

Ayan Barua, co-founder and CEO, learned the cost at Siftery and G2, where integrations pulled engineers from the product. Lauren Long, co-founder and CTO, met it while building Firebase Extensions at Google. They interviewed 100+ companies before writing code.

An agent can know that an invoice is overdue, a lead needs follow-up, or a record needs an update. It still has to find the right object, field, permission, credential, and approval rule inside the customer's system.

One bad read makes the demo awkward. One bad write can damage the pilot or renewal.

Ampersand gives developers 5 primitives: Read, Subscribe, Search, Write, and Proxy. Teams define integrations in version-controlled configuration. Its runtime handles customer schemas, authentication, token refreshes, retries, backfills, monitoring, and mapping after launch. Customer admins map fields inside the onboarding flow, so the integration ships with the product instead of after it.

That “after launch” part carries the business case. Ampersand says one customer-specific Salesforce implementation can consume 1–2 months of engineering work. Bessemer reports that 90% of completed proofs of concept become paying customers. The first figure is company-reported; the 90% figure is investor-reported by the round's lead, but they show why integration depth can decide which enterprise deals reach production.

Ampersand says 11x, Orb, and Square use the platform. Hatch CTO John Pena said it opened customers previously blocked by complex integration requirements.

Bessemer led the round. Existing investors Matrix and Flex Capital participated; Yelp, Tenacity Capital, CTO Fund, Mana Ventures, and angels also joined. Bessemer partner Lauri Moore is taking a board seat. Ampersand says total funding is now $20M.

The money will deepen ERP and industry-specific coverage, build Andi, Ampersand's AI integration agent, and expand a forward-deployed bench of people who understand systems such as NetSuite, SAP, Workday, ServiceNow, Oracle, and Epic.

Enterprise customization is not disappearing. Those fields and workflows encode how a company sells, bills, approves, and operates. Ampersand is building for that stubborn reality without turning every signed customer into another private integration project.

Stuut

An enterprise can finish the sale and still watch the cash stall behind a missing purchase order, an invoice sent to the wrong inbox, or a portal nobody logged into.

That gap turns accounts receivable into a detective agency with spreadsheets. Finance searches for the document, sales protects the relationship, and operations checks what shipped until the payment lands.

Stuut has raised a $52.5M Series B to take more of that investigation off the humans. Insight Partners led the round, with Andreessen Horowitz and M12, Microsoft's Venture Fund, participating. Stuut says the financing brings total funding to $93M.

Tarek Alaruri, co-founder and CEO, and Ben Winter, co-founder and COO, started with collections. The company now wants its AI coworker to carry order management, credit, collections, payments, cash application, disputes, and deductions. CTO Mek Stittri is scaling the foundation underneath that expansion.

Stuut's agents operate across email, SMS, voice, ERPs, bank accounts, CRMs, and payment systems. The company says 81.7% of outbound collections activity runs without human involvement and 95% of incoming payments are matched automatically. Those are company-reported metrics, but they describe the standard enterprise buyers will use: execution with an audit trail instead of another dashboard waiting for attention.

Stuut reports more than 150 customers, 5x customer growth in a year, and more than $3B moved through the platform. Honeywell, ZoomInfo, and Verifone appear in its current materials. These buyers want fewer days between revenue booked and cash available.

The financing record needs careful accounting. Stuut announced a $52.5M Series B, while a September SEC Form D reports $67.6M of equity sold to 11 investors without naming the round or participants. Public sources do not reconcile the figures.

Insight Partners is backing a larger claim than faster collections. If Stuut can preserve context from the purchase order through bank reconciliation, the agent becomes part of the financial infrastructure surrounding every sale. That creates a deeper product and a much higher trust threshold.

The next stage will be decided inside the exceptions: the disputed deduction, wrong entity, missing proof of delivery, and customer relationship that cannot be handled like a generic overdue balance. Stuut now has the capital to follow those transactions until the software earns the right to move the cash.

Teddy AI

The first thing Teddy AI asks a law firm is not how many hours it bills right now.

It asks how much revenue lives with the top 3 clients, how much lives with the top 3 lawyers, how many clients and lawyers leave each year, and how much revenue returns without being resold from zero.

Those questions explain the company better than the .AI in its name.

Teddy AI has raised a $60M Seed round after reporting more than $25M in B2B revenue during 2026. Tucker's Farm Corporation incubated the compliance- and client-oriented legal services platform.

Kyle Tucker, founder of Tucker's Farm's private-equity arm, said a roughly 3-week process drew about $115M of equity interest before Teddy closed $60M. The confidential investor group includes traditional LPs, an endowment manager, and repeat partners.

Teddy's website is built around valuing and partnering with law firms. Its questions find the parts of a professional-services business that can look healthy in a revenue report and fragile in operation.

A firm can have excellent attorneys and still depend on 3 rainmakers. It can serve valuable clients and still watch too much revenue walk when one relationship changes. AI can improve speed while creating new questions about confidentiality, supervision, pricing, and accountability.

That makes Teddy's language disciplined for a company with AI in the name. Its stated mission is compliance and client service, with technology used when it improves those outcomes. The harder job is keeping trust, knowledge, service quality, and operating discipline intact while the tools change.

Thomson Reuters reports that 77% of clients consider AI-enabled quality improvements very important or essential, while only 5% say most or all outside providers deliver them. Clients want better economics and better execution while firms decide what AI strengthens and what it exposes.

Teddy AI has not disclosed its leadership, exact legal niche, customers, investor names, valuation, or use of funds. Its $25M revenue figure is company-reported. This round funds a platform before the public can see the operators, portfolio, or controls carrying it.

The valuation form shows where Teddy plans to work. Client concentration, lawyer concentration, attrition, recurring revenue, compliance, and infrastructure are where a legal practice becomes an institution instead of a collection of relationships waiting for one departure to redraw the economics.

Mecka

Physical AI has a supply chain hiding inside the word intelligence: somebody still has to manufacture the experience.

A language model can inherit a library. A robot gets a factory floor, kitchen, lab, or warehouse where motion and the human workarounds were never written down. Machines need examples of how work actually happens before they can become useful inside it.

Mecka is building that missing production system. The company has raised a $60M Series B led by Sequoia Capital, with NVIDIA, M12, Qualcomm Ventures, and Samsung joining as new investors. Kindred Ventures, Framework Ventures, and Neo returned, alongside angels including Tony Xu, Frank Slootman, and Milan Kovac.

The founding team came from payments, software, and operations. Josh Gao, CEO, Jason Chong, CTO, Mogen Cheng, and Duy Nguyen built Mecka around a costly premise: human activity can become machine-training infrastructure if someone can capture, align, qualify, and deliver it at industrial scale.

That is the business. Mecka designs multi-sensor capture hardware, runs collection fleets, and uses computer vision and multimodal models to turn messy demonstrations into structured signals. EgoVerse connected the company with researchers at Georgia Tech, Stanford, UC San Diego, ETH Zürich, MIT, and Meta to study human-to-robot transfer across labs and machines.

Now Mecka is following the data into deployment. Enterprise customers want physical AI but may lack a robotics team capable of connecting hardware, post-training, site integration, and ongoing operations. The Series B expands that handoff with more instruments, deeper research, larger data infrastructure, and commercial deployments.

The strategic group reflects that shift. Sequoia brings the venture bet. NVIDIA, Microsoft, Qualcomm, and Samsung sit across compute, platforms, chips, devices, and the infrastructure surrounding intelligent machines.

Mecka says it surpassed $100M in run-rate revenue in June and expects $300M by year-end. Those figures are company-reported, most customers remain unnamed, and the projection is still ahead. The speed shows how urgently robotics teams and enterprises are buying experience they cannot scrape from the internet.

The capital now follows Mecka to the customer site. Each deployment will produce new edge cases, demonstrations, and evidence about whether the system improves after installation. That loop is where a data vendor becomes operating infrastructure and physical AI learns from the places it is supposed to work.

Atomic Machines

A factory usually remembers one product in steel, tooling, masks, fixtures, and process lore. Atomic Machines wants the instructions to live in code instead.

Jeff Holden spent 6 years building that bet in stealth. Atomic Machines has now emerged with $250M raised to date, the Matter Compiler, and a first product aimed at one of the more violent timing problems inside AI infrastructure.

The Matter Compiler is an AI-native manufacturing system for functional micro-machines. It builds moving, multi-material devices with single-digit-micron features, measures each operation, and corrects the next one in real time. New design, new instructions. No product-specific tooling.

That sounds clean because software trained all of us to expect clean. Matter has a union contract with physics.

Atomic Machines made the argument concrete with PrimeSwitch PS-150, a surface-mount electromechanical relay rated for 150 amps. The company says it opens in 50 microseconds with 200 micro-ohms of on-resistance, 1,500 volts of galvanic isolation, and zero holding power.

Those numbers matter as AI data centers move toward 800-volt DC power. Mechanical contactors conduct efficiently but open in milliseconds and can arc. Solid-state switches react in microseconds, yet produce heat and still need mechanical isolation. PrimeSwitch splits the job so silicon interrupts while metal carries and isolates.

Founder and CEO Jeff Holden helped build Amazon Prime and later served as Uber's first chief product officer. CTO Klaus Zietlow brings surgical-robotics and medical-device experience. Backers include OneIM, Gigafund, XTX Ventures, KAS Venture Partners, Valor Equity Partners, and others.

The financing needs one accounting correction: Atomic Machines disclosed $250M raised to date, not a named new $250M round. It did not disclose a lead investor, valuation, or how the capital was divided among prior financings.

PrimeSwitch is shipping to unnamed early-access customers for evaluation. That is where the thesis leaves the presentation and meets yield, reliability, qualification, cost, and the customer engineer who has to trust a tiny relay with a very large electrical consequence.

If one factory can genuinely move between device designs without rebuilding the process around each product, Atomic Machines is not merely shrinking machinery. It is trying to change where manufacturing knowledge lives, and PrimeSwitch is the first sentence customers get to grade.

Rein Security

The dangerous part of an enterprise AI agent often begins after the prompt looks ordinary.

An onboarding agent opens a PDF. A shopping assistant receives a normal request. Then the software calls an API, reaches into a system, follows an instruction nobody intended, or probes infrastructure at machine speed. The risk has already crossed into production.

That handoff is where Rein Security is placing its bet.

Matan Bar-Efrat, co-founder and CEO, and Netanel Rubin, co-founder and CTO, built Rein around runtime context: observe what an agent executes, which resources it touches, and whether the action should be allowed. Rein describes its platform as a patented sidecar that applies real-time guardrails while keeping data inside the customer's environment.

Rein has raised a $25M Series A co-led by Glilot Capital and Sienna Venture Capital, with Corner Ventures, Atlacle, and RNP Capital Advisors participating. The company says the round brings total funding to $35M and will finance product work, agent-security research, and global hiring.

The capital arrives behind early operating evidence, with a qualifier. Rein reports 8x revenue growth and 5x customer growth since its January 2026 launch, plus thousands of protected agents executing millions of actions. Flex, Swimlane, Lemonade, and Dun & Bradstreet are named users. The absolute revenue, customer counts, and contract values are not public.

The funding math deserves precision. Rein disclosed an $8M seed in January and a $25M Series A now. Those rounds total $33M, while the company reports $35M raised to date. The reviewed announcements do not identify the remaining $2M.

Investors are backing a category decision as much as a company. Gartner expects the market for securing AI to reach about $4.8B in 2027, up 68.7% from 2026. That opportunity will attract gateways, governance platforms, application-security vendors, and every incumbent capable of adding an AI tab to a dashboard.

Rein's argument is more specific: prompt inspection is too early and post-incident review is too late. Security has to travel with the agent into the moment where language becomes access, code, and business action.

The $25M gives Rein more room to prove that runtime control can become a durable enterprise budget line. The buyer's standard will be unforgiving: see enough to stop harmful action, preserve the context needed to explain it, and stay light enough that production teams will keep the control turned on.

Peppermint

Clinical research treats protocol deviations like emergencies, then lets the financial record live across contracts, visit logs, invoices, remittances, and spreadsheets that may never agree.

Peppermint raised a $4.7M Seed round led by Moxxie Ventures, with Homebrew, Better Tomorrow Ventures, Layout Ventures, Digital Health Venture Partners, Night Capital, Benjamin Mann, and Evan Moore participating.

Peppermint was founded by David Freeman, CEO, and VJ Thurimella, CTO. They met at X1 as its first business hire and first engineer, then helped build payments and reconciliation infrastructure for the credit-card business Robinhood acquired and used to launch Robinhood Gold Card.

A clinical-trial site can complete the visit, document the work, and satisfy the protocol while its revenue remains stuck behind a contract term, an unbilled procedure, a sponsor portal or a payment that arrives without enough detail to reconcile it.

Peppermint connects to the clinical trial management and accounting systems a site already uses. Its AI reads visit activity, contracts, and payment history, while human billing specialists handle the judgment calls. The platform validates billable work, prepares invoices continuously, matches payments to visits, and supports collections without forcing a replacement of the research team's operating stack.

Peppermint reports that partners collect revenue 3x faster, have cut overdue invoices threefold, and reclaimed 20% of staff time previously spent on billing. Studies under management have tripled in three months. Those are company-reported results, not an independent audit, but they point directly at the pressure this round is financing.

The next product move is bigger than collections. Peppermint plans to expand into budget negotiation, expense management, and clinical-trial analytics, creating a ledger that can show what a study earns, what it costs and where margin disappears across a portfolio.

That is also where the work becomes less forgiving. Clinical finance has too many exceptions for software that is merely fast. Peppermint has to preserve the evidence behind every invoice and payment well enough for sites, sponsors, and CROs to trust the ledger when the numbers disagree.

Research sites are expected to protect patients, execute protocols, and produce reliable data. Peppermint's $4.7M gives the team room to make financial precision part of that research capacity, with every recovered dollar buying back time for the trials themselves.

VEIR

A data center architect has to leave room for the electricity before the next generation of AI racks can move in. Copper takes up space, installation takes labor, and a denser compute plan asks more of the building around it.

VEIR is taking that problem into manufacturing. The Woburn, Massachusetts company closed a $110M Series C co-led by Matter Venture Partners and Tyche Partners, bringing total capital raised to $225M. The money supports manufacturing capacity, its product portfolio, and the technical and commercial teams needed for deployment.

CEO and co-founder Tim Heidel and CTO Erhan Karaca are building an integrated power delivery system: superconducting cables, cryogenic cooling, controls, termination units, and monitoring. The customer has to buy a system that can be installed and operated inside a critical facility. A remarkable cable still needs the rest of the equipment to cooperate.

In November 2025, VEIR reported delivering 3 MW through a single low-voltage cable in a simulated data center environment. That gives the engineering conversation something concrete. The company is now working with customers and industry partners on AI-factory reference designs while preparing initial commercial deployments.

There is a useful change in the buying conversation here. Electrical routing, physical footprint, materials, and installation effort become part of the compute-capacity decision. Engineers designing the campus need a seat early enough to influence what the campus can support.

The syndicate spans new participants LG Technology Ventures, Gates Frontier, Sabanci Climate Ventures, and Hui Capital, alongside returning Engine Ventures, Galvanize Climate Solutions, Piva Capital, Congruent Ventures, and VXI Capital, among others. VEIR previously closed a $75M Series B led by Munich Re Ventures in January 2025.

Co-founder Steve Ashworth developed the early cooling concept; Heidel's grid work reaches back through MIT, ARPA-E, and Breakthrough Energy Ventures. AI data centers give that power-delivery experience a demanding near-term customer: an operator trying to fit more electrical capacity into a site with its own geometry, construction sequence, and uptime obligations.

The Series C funds the manufacturing and deployment infrastructure around that customer. As reference designs move toward installations, VEIR's engineers will be carrying the conversation from how much power a cable can move to how an entire facility makes room for it.

Bessemer Venture Partners

At Bessemer Venture Partners, a partner can back a founder without winning an internal election. The firm operates without a CEO or central investment committee, giving its investors the authority, and the responsibility, to make their own calls.

That independence now sits alongside $5.75B in new capital. Announced September 23, the single close puts $1.75B toward seed and early-stage investing and $4B toward growth. Two different moments in a company's life, with a much larger pool reserved for the later one.

Jeremy Levine, Partner, describes the continuity: study a technology shift early, find the founders pursuing it, and keep investing as their ambitions expand. Byron Deeter, Partner, connects the growth practice to the speed of AI-native companies, with dedicated capital and partners able to lead concentrated rounds.

The distinction matters to a founder who has outgrown the check that first got the company moving. A useful early investor may know the customers, the failed experiments and the people doing the work. Financing the next stage still requires enough capital, a fresh judgment and a willingness to take a larger position.

Bessemer's growth platform can invest in existing portfolio companies and businesses it is meeting for the first time. That second group matters. Institutional memory can help an investor understand a company, but the mandate also leaves room for founders whose first check came from somewhere else.

The firm says roughly 70% of its investments remain early-stage. Deal count and capital allocation tell different stories: $4B of this close is for growth. Bessemer is maintaining a broad search for young companies while giving its later-stage partners more room to finance the ones they choose.

Its experience spans software companies such as Toast and ServiceTitan, AI businesses including EliseAI and Legora, and physical systems such as Waymo. Those relationships are part of the knowledge Bessemer brings to a new investment. The announcement does not assign those companies money from these funds.

For founders, the useful conversation gets more specific than whether an investor likes AI. Who can make the decision? Which stage can the firm finance? And will the partner who understands the business still have a seat in the discussion when the round gets larger?

Bessemer has funded both sides of that discussion. Its partners will be having it with founders they have known for years and others walking through the door for the first time.

Vesta

Vesta can put an AI agent on a mortgage file. Getting a lender onto Vesta can still take most of a year.

That is the next piece of work behind its $30M Series B, led by Conversion Capital. Co-founder and CEO Mike Yu and co-founder and CTO Devon Yang have already rebuilt the core system. Now they have to help lenders move their operating knowledge into it faster.

The San Francisco company started in 2020, after the founders worked at Blend. They spent 2 years building the platform before converting lenders. A mortgage system carries a business's rules, exceptions, handoffs, and history. Replacing it means asking an institution to change how money moves through the building.

Pennymac, New American Funding, and nbkc bank are investing in this round alongside Citi Ventures, First American, FirstKey Mortgage, a16z, Zigg, and Navitas. Total funding reaches $85M. Some of the people evaluating the software are also financing the company responsible for making it work.

Pennymac's SEC-filed Q4 2025 presentation gives that relationship some useful numbers: roughly 50% efficiency gains for loan officers and about 25% less loan processing time. Pennymac also holds a minority equity stake. Those are a customer's reported implementation results, with its financial interest disclosed.

Vesta says revenue grew more than 12x over the past 12 months. Lenders live on the platform or moving to it originate more than $100B annually; that figure includes migrations still ahead. About 40% of completed tasks now run through AI agents and automated workflows together.

Inside the product, people and agents share the loan file. Teams set permissions, review the work and take the exceptions. The system records actions and reasoning, giving a reviewer somewhere to start when a decision needs an explanation.

Yu's next ambition reaches into implementation itself. Today, teams translate lender procedures into thousands of settings manually. Vesta wants an agent to read those procedures, ask questions, and work through old loans so the configuration can take shape faster. That remains a development direction, with plenty of lender-specific judgment attached.

For mortgage operators, the interesting handoff is moving. Once the routine work has somewhere to run, the experienced person gets a different assignment: deciding which instructions the system should follow, where it must stop, and how the next loan teaches the organization something useful.

Waymo

Steve Fieler's job at Waymo now includes a relationship the company has never had before: lenders. The CFO is helping finance a driverless service whose next stretch of growth comes with a repayment obligation.

Waymo closed its first debt financing on October 8: a $5B term loan with PIMCO, Blackstone, and Sixth Street as lead syndicated lenders. Capital Group, Loomis Sayles, and T. Rowe Price joined as significant lenders. Goldman Sachs served as sole lead bookrunner. Apollo, Blue Owl, Diameter Capital Partners, Franklin Templeton, Fidelity Management & Research Company, HPS Investment Partners, and Oaktree also participated.

The company already raised $16B in equity in February at a $126B post-money valuation, with Alphabet remaining its majority investor. That financing and this loan do different jobs on the capital structure. The October announcement leaves the interest rate, maturity, and security terms undisclosed.

Fieler previously held senior finance responsibilities at Alphabet and served as HP's CFO. That background now sits alongside co-CEOs Tekedra Mawakana, focused on commercialization, and Dmitri Dolgov, focused on developing and deploying the Waymo Driver. Dolgov helped found the Google self-driving project in 2009, before it became Waymo in 2016.

Their operating calendar stretches from rides people can take today to cities still preparing for a launch. Waymo says it opened service in its 15th U.S. city last month. In Las Vegas, public access began rolling out in September with Ojai vehicles powered by the 6th-generation Driver.

Singapore has a longer runway. Waymo is working with the Ministry of Transport and Land Transport Authority toward intended commercial service in 2028, with manual readiness driving planned for 2027. Local road geometry, monsoon conditions, and regulatory approvals all belong to that expansion story.

A software improvement can travel quickly. A ride-hailing operation still needs a place, a fleet, and permission to put passengers in it. Credit gives Waymo another way to finance that physical buildout while the company carries several deployment calendars at once.

The lenders are joining a business that has moved far beyond the first demonstration ride. Its economics still require their own evidence; this announcement supplies no standalone profit figure. For Mawakana, Dolgov, and Fieler, more financial flexibility now has to meet the work of bringing each city into service, including the ones where the first passenger is still years away.