The Work They Saw Anyway

Pilgrim

A biological threat does not wait for a courier.

Pilgrim is betting that early warning should begin where the air is moving, not days later when a sample reaches a centralized lab.

The Redwood City defense-biotechnology company raised a $25M Seed round at a reported $150M valuation, led by Buckley Ventures. The Wall Street Journal reports that Anthropic employees Logan Graham and Sholto Douglas invested personally. People close to frontier AI risk are backing a physical layer of biosecurity.

Founder and CEO Jake Adler is building Pilgrim around that physical layer. ARGUS is a 50-pound autonomous biosurveillance system designed to collect airborne biological material, screen for known and novel threats, and analyze results on-device. Pilgrim lists a time to result under 60 minutes, with UV LIF, qPCR, and sequencing detection modes.

The United States can produce remarkable biological research, but early warning still depends on where samples are collected, how quickly they move, and whether anyone can act before exposure becomes an outbreak.

Pilgrim says ARGUS is meant for defense, homeland security, and public health. The company also says it has signed an agreement with the CDC for biosurveillance, opened a 20,000-square-foot Redwood City facility, and begun federal deployments. Those are company-reported operating signals, not a substitute for independent field-performance data.

That distinction matters at a $150M Seed valuation. A detector is useful only when the institution around it knows what happens after the alert. False positives have costs. Missed detections have consequences. Genomic data needs secure handling. Field reliability, manufacturing, confirmatory testing, response protocols, and procurement all have to work together.

The capital gives Pilgrim room to prove that ARGUS can become infrastructure instead of an impressive box. Buckley Ventures is underwriting a possible shift from episodic sampling toward persistent biological awareness at bases, airports, hospitals, ports, farms, and other high-risk environments.

Pilgrim's earlier $4.3M round helped bring battlefield medicine and biosurveillance out of stealth. This $25M round changes the obligation. Jake Adler and the team now have to show that faster detection can survive real air, real institutions, and the chain of decisions that begins after a machine says something dangerous may be present.

That chain is where a biology prime gets built.

Deeper Analysis 👇

https://devcuration.com/articles/pilgrim-raises-25m-argus-biothreat-detection

Sol

The expensive part of an email promise is rarely typing the sentence. It is remembering documents, people, calendars, and follow-ups that sentence quietly puts in motion.

“I’ll send the analysis” looks finished in the inbox. Operationally, it just created research, a draft, an attachment, and a waiting conversation.

Sol Foundry Inc. has raised $4M to build around that gap. The company emerged from stealth with backing from General Catalyst, Nexus Venture Partners, DeVC, peercheque, and Kunal Shah.

CEO and co-founder Anish Karan built Sol with co-founders Prateek Srivastava and Ranjith Nair. Sol scans Gmail for commitments, gathers context, and begins the supporting work: research, documents, presentations, meeting coordination, or a draft reply.

Sol prepares the work, but the user still approves anything before it is sent, scheduled, or shared. That boundary matters when an assistant is reading the language people use with customers, candidates, colleagues, and investors. Proactivity is useful only if judgment stays visible.

Sol says the product runs in its computer environment, can use a browser, and draws on more than 100 specialist skills. More than 100 people are using the pre-launch platform, according to the company, while access remains selective.

The $4M will support R&D, model-processing and tooling costs, hiring, customer work across geographies, and go-to-market expansion. Most near-term hiring is expected in the US.

The inbox is already a rough system of record for commitments, but most AI products still make the user translate those commitments into new instructions. Sol is betting that the next useful interface is not another empty prompt box. It is the context people have already created while doing business.

That creates a harder engineering and trust problem than drafting a clever reply. The system has to identify a real commitment, distinguish it from casual language, assemble the right context, choose the right tools, and prepare work without creating a mess the user has to untangle.

Sol’s security page says customer data is not sold or used to train models. It also reports a CASA Tier 2 assessment and a SOC 2 Type I readiness assessment, which the company distinguishes from a formal SOC 2 audit report or certification.

The capital gives Sol room to prove whether proactive AI can become dependable infrastructure for commitment-heavy work. The value will show up in the small promises that stop disappearing between an email thread and the workday built around it.

Deeper Analysis 👇 https://devcuration.com/articles/sol-raises-4m-email-commitments-ai

Bird

The risky part of an AI agent starts after it writes a perfectly reasonable sentence.

Now it wants to send the email, follow up on WhatsApp, place the call, and keep moving while nobody is watching. That is where a model meets identity, consent, compliance, delivery, and a customer who never volunteered for a software demonstration.

Bird has spent 15 years building the infrastructure behind those moments. Robert Vis and the founding team started MessageBird in 2011, long before “agentic” became a line item in every software roadmap.

Now Bird has completed $450M of debt financing while opening that infrastructure to AI agents through Agentic Harness. J.P. Morgan led the package, with Capital One, Citi, Silicon Valley Bank, MUFG, Flagstar, and Huntington in the lender group.

The capital deserves accurate accounting. This is a $400M term loan plus a $50M revolving credit facility, structured as a dividend recapitalization. Bird says it will provide liquidity to existing shareholders, including current and former employees, while the company stays private and independent.

That differs from raising $450M of equity for a product roadmap. Shareholders get cash without a public offering or new equity price. Bird gets flexibility and a debt obligation whose rate, maturity, covenants, and leverage were not disclosed.

Agentic Harness is the sharper operating story. It gives agents access to Bird through MCP, CLI, and skills to manage email, WhatsApp, calls, and the resources behind those interactions.

Generating language is cheap. Giving software permission to communicate as a business is where the bill, the regulation, and the reputation arrive.

Bird believes its existing network is the advantage. The company says it moves trillions of messages annually across more than 150 countries. It also reports $165M of EBITDA in 2025 after extensive automation and says headcount fell from more than 1,000 at peak to 120.

Those figures are company-reported, and the smaller team raises questions about resilience, support, and institutional memory. The lenders are underwriting a thesis: communications infrastructure can become an execution layer for agents.

The next wave of AI infrastructure will be judged less by how confidently a model speaks and more by what happens after someone lets it act. Bird now has a bank syndicate, shareholder liquidity, and 15 years of communications machinery riding on that handoff.

Deeper Analysis 👇 https://devcuration.com/articles/bird-com-receives-450m-debt-financing

AcuityMD

The sales rep knows a surgeon is changing hospitals. Finance knows a contract moved. Market access knows reimbursement shifted. The CRM knows whatever someone remembered to type last Friday.

That is the commercial reality waiting behind every cheerful promise that AI will transform MedTech.

AcuityMD raised an $80M Series C led by StepStone Group, with Benchmark, Redpoint Ventures, ICONIQ, Atreides Management, HighSage Ventures, Artisanal Ventures, and others participating. The April round valued the Boston company at $955M and brought total funding above $160M.

Founded in 2019 by Michael Monovoukas, Lee Smith, and Robert Coe, AcuityMD tackles a stubborn problem: MedTech commercialization runs on context scattered across claims, FDA records, reimbursement, physician behavior, facilities, territories, contracts, and private sales history.

Horizontal AI can write a convincing answer about that market. Convincing and useful are not the same billable event.

AcuityMD built a MedTech ontology mapping physicians, facilities, networks, procedures, reimbursement, and their relationships. AcuityAI adds the customer's commercial reality and the rep's situation so an answer can become a target list, account plan, territory decision, or next action.

That distinction explains the investor logic. StepStone Group is not backing a chatbot with better bedside vocabulary. The wager is that vertical context gains value as general models improve, because the model still needs to know which surgeon performs the procedure, where the patient moves, what the contract permits, and why yesterday's account plan is stale.

AcuityMD says more than 500 MedTech companies use the platform, including 16 of the top 20, and that customers have identified more than $34B in pipeline. Those are company-reported figures, but they show the scale of the commercial surface AcuityMD wants its agents to own.

The capital will push 3 jobs forward: expand agentic AI for sales, leadership, and marketing; deepen the ontology; and move beyond the commercial organization toward the full MedTech product lifecycle.

That expansion is where the story gets difficult. A system influencing portfolio strategy, product development, reimbursement, or launch sequencing carries wider consequences and a longer list of people who must trust the underlying data.

Michael Monovoukas, Robert Coe, and Lee Smith have spent 7 years turning MedTech's scattered market memory into software. The $80M buys them room to turn that memory into action while the rest of enterprise AI learns that fluency is cheap and context still has to be earned.

Rightway

Benefits leaders buy pharmacy savings through contracts most employees will never read. The incentive inside that contract can matter more than the discount printed up front.

That is the purchasing problem Rightway is financing with a $155M Series E led by Francisco Partners, with Thrive Capital and Khosla Ventures returning.

The traditional PBM pitch starts with discounts and rebates. The harder question is who earns more when the drug costs more, who owns the dispensing channel, and whether the employer can follow the money to member checkout.

Rightway says its answer is one administrative fee per member, every rebate dollar passed through, no spread or dispensing revenue, and no pharmacy ownership. SureSpend caps total pharmacy spend, while its Zero-Markup Wrap covers GLP-1s and rare high-cost medications at net cost.

That model grew out of a human problem. Co-founder and CEO Jordan Feldman built Rightway with his father, cardiologist Dr. Theodore Feldman, around the guidance people usually get only when someone in the family understands healthcare.

Care navigation could help a member find a specialist or make sense of a bill. It could not fully change medication costs while another company controlled the formulary, pricing, and pharmacy channels. Rightway built the PBM because guidance without control left the most expensive decisions elsewhere.

The company says 45 Fortune 500 businesses now run pharmacy benefits on its model. That figure is company-reported, but the commercial signal is clear: large employers will reconsider incumbent PBM structures when an alternative can make the economics legible and still support members.

The capital will expand the AI and technology behind the pharmacy benefit. Rightway plans to remove administrative work from pharmacists, surface lower-cost and higher-value options earlier, and give clinical teams better context for personalized support.

Chief Pharmacy Officer Kristin Devlin, PharmD, put the human side plainly. Pharmacists entered healthcare to help people, yet retail and administrative systems can leave too little time for that work. The useful role for AI here is to return judgment to the pharmacist, not turn the pharmacist into another software field.

The round gives Rightway more capacity and employers a larger system to audit. Every formulary choice, high-cost prescription, rebate reconciliation, and support call will show whether aligned incentives stay visible as the company and the drug budget grow.

Deeper Analysis 👇 https://devcuration.com/articles/rightway-raises-155m-series-e-pharmacy-benefits-ai

Basalt Health

Healthcare built a high-stakes admissions market on a low-bandwidth ritual: fax the chart, wait, and hope the right bed answers before someone else does.

A patient, nurse, discharge team, and available bed can run on different clocks while the connecting information sits inside a referral packet.

Basalt Health has raised a $20M Series A led by returning investor New Enterprise Associates, with Frist Cressey Ventures and 25m Health participating. The Nashville company says the round brings its total capital raised to approximately $24.5M.

Founder and CEO Ben Hackett started Basalt in 2024 after nearly a decade at Accolade. The company focuses on post-acute admissions, where a facility must decide whether it can safely care for a patient before the hospital, payer, or bed moves on.

Basalt reads the referral when it arrives by fax, portal, or direct integration. It checks the record against the facility's clinical and payer criteria, then surfaces red, yellow, or green findings with citations back to the exact source.

That last part carries the business. Healthcare AI can make a recommendation quickly. A clinician still has to understand the evidence, own the decision, and defend it when the admission, authorization, or appeal becomes complicated.

Basalt says its Lifepoint Health deployment cut median chart-processing time from 8.5 minutes to 1.2 minutes, an 86% decrease. It is expanding across 49 more Lifepoint markets and 62 ScionHealth hospitals by the end of 2026, with the latter covering more than 6,000 practitioners.

Those are company-reported results, not an audited victory lap. They are still the kind of operating evidence enterprise healthcare buyers want before allowing new software into a live patient handoff.

The round will fund expansion into discharge and payer workflows, plus hiring across engineering, implementation, sales, and marketing. That is the work after the benchmark: integration, local criteria, security review, training, support, and enough trust to stay in the workflow.

The investors are backing a specific healthcare AI bet: products that disappear into the operation while making every important reason easier to see. That is a harder promise than a clean chatbot demo.

Basalt now has to carry that evidence discipline across more systems without letting speed outrun accountability. Every new facility will bring its own criteria, people, payer logic, and patients waiting for the next bed.

Deeper Analysis 👇 https://devcuration.com/articles/basalt-health-raises-20m-series-a-led-by-nea

Prime Minute

Wildfire technology can spot a problem from space. Getting 110 gallons of suppressant onto the right patch of ground at night, through smoke and wind, is where the market stops sounding like software.

Prime Minute emerged from stealth with $15M in pre-seed funding led by Eclipse, with 8VC participating. The company is combining real-time intelligence, AI-assisted mission analysis, and precision aerial suppression, starting with wildfire.

Its first hardware expression is a Precision Guided Suppression Unit, or PGSU. Each roughly 1,000-pound unit carries 110 gallons of water and uses GPS guidance plus onboard flight controls to steer toward a predetermined target. A dispersal system is designed to spread the payload across a 35-foot suppression radius.

That is company and testing-partner language, not a commercial victory lap. Colorado's Center of Excellence for Advanced Technology Aerial Firefighting says the PGSU remains in operational testing. Phase 1 focused on safety, accuracy, and precision. Later work must answer the harder question: how well does the system suppress fire, and how many units are needed for an emerging ignition?

That boundary is why the round matters. Fire agencies have invested in detection, mapping, forecasting, aircraft, crews, and command systems. Prime Minute is attacking the handoff between knowing and acting when darkness, smoke, weather, terrain, or scarce resources keep traditional aerial options away.

Ronen Korman, Itai Mendelsohn, Gabe Sganga, and Joey F. are identified as cofounders by Eclipse. Gabe Sganga is chief growth officer; Inc. reports Lior Susan is cofounder and executive chairman.

The $15M will fund technology development, local manufacturing, team expansion, testing, and validation. Prime Minute plans to sell suppression as a service rather than the units, shifting the buyer conversation from ownership to readiness, response capacity, and protection.

That model has an obvious customer set and an unforgiving proof requirement. Governments, utilities, insurers, and data-center operators carry wildfire exposure. They cannot underwrite the promise on a slick demo. They need safety data, operating protocols, aircraft integration, procurement clarity, and evidence that precision delivery changes outcomes.

Prime Minute is trying to make the first hour of a wildfire an engineered operating window instead of a period everyone documents while the fire gets bigger. The capital buys more chances to test whether that window can be used.

Deeper Analysis 👇 https://devcuration.com/articles/prime-minute-raises-15m-wildfire-response

Precision Neuroscience

Brain-computer interfaces are usually photographed as hardware. The expensive part sits outside the frame: clinical protocols, FDA files, clean-room manufacturing, hospital integration, and surgeon trust.

Precision Neuroscience closed an oversubscribed $250M Series D co-led by Pershing Square Inc., the Ackman Oxman Institute, and an undisclosed life-sciences fund. Duquesne Family Office, B Capital, ARK Invest, Invus, Mubadala Capital, Mirae Asset Capital, Korea Investment Partners, Hitachi Ventures, and JSL Health Capital also participated.

The round brings Precision's company-reported total capital to $430M. It will fund clinical expansion, additional FDA work for Layer 7, and commercialization infrastructure. The valuation and investor check sizes remain private.

Layer 7 is a thin, flexible array that rests on the brain's surface instead of penetrating tissue. Each module carries 1,024 electrodes. Precision has demonstrated recording across 4,096 electrodes and real-time decoding of intended movement.

The regulatory line matters. FDA clearance covers the Layer 7-T cortical electrode for recording, monitoring, and stimulation for up to 30 days. Precision's fully implantable wireless BCI remains investigational and unavailable for sale in the United States.

Precision reports more than 100 patient procedures across 18 medical institutions. It has a development partnership with Medtronic and owns a MEMS manufacturing facility in Texas. Together, those assets shorten the distance between a neural signal recorded in a study and a device a hospital can use repeatedly.

Michael Mager and Benjamin Rapoport founded Precision around a surface-based approach to brain access. The original team also included engineers Mark Hettick and Demetrios Papageorgiou. Michael Mager is CEO, Benjamin Rapoport is Chief Science Officer, Mark Hettick is VP of Engineering, and Nate Pletcher is CTO.

The syndicate spans public markets, family offices, venture firms, sovereign capital, and a new brain institute. It is a long bet on clinical evidence, manufacturing discipline, regulatory execution, and hospital access.

Precision can now push those dependencies in parallel. The next evidence will come from permanent-implant review, repeatable outcomes for intended users, production quality, and whether hospitals can absorb the technology without turning every deployment into a custom research project.

The electrode reads the signal. Precision still has to build everything that lets medicine trust it.

Deeper Analysis 👇 https://devcuration.com/articles/precision-neuroscience-raises-250m-bci-scale

PicoJool

The most expensive processor in an AI cluster can spend part of its day waiting. Not for a better model. For data to arrive from the processor beside it.

That queue is where PicoJool is putting a $27.5M Series A.

Founder and CEO Al Yuen built PicoJool around vertical-cavity surface-emitting lasers, better known as VCSELs, and the optical links that move data across AI infrastructure. Socratic Partners led the round, Hudson River Trading participated, and the financing follows a $12M seed led by Playground Global. PicoJool now reports $39.5M in total funding.

The check is paying for a handoff that semiconductor announcements tend to compress into one polite sentence: from “the performance is there” to customers can qualify it, foundries can produce it, modules can ship it, and support teams can stand behind it.

PicoJool says its portfolio includes 100G and 200G VCSEL products plus 50G NRZ and 64G NRZ/PAM4 microVCSEL configurations. It targets 800G, 1.6T, and 3.2T links across active optical cables, near-packaged optics, and co-packaged optics. Its 200G products exceed 37GHz of bandwidth, according to the company.

Those specifications matter because AI infrastructure has an expensive coordination problem. Adding accelerators increases theoretical compute. Keeping them useful requires a network that can feed data between processors without letting bandwidth, power, and cost eat the gain.

That is also why the investor roster makes sense. Socratic Partners is built around semiconductor operators. Hudson River Trading knows what latency and compute economics feel like when small inefficiencies become real money. Playground Global backed the $12M seed.

Now PicoJool has to turn device performance into repeatable infrastructure. The company has started early chip-level sampling and is working with WIN Semiconductors and other GaAs foundries to prepare its 200G VCSELs for production. The Series A will expand teams and facilities in the U.S. and Taiwan, customer qualification, manufacturing, sales, and support.

The company has not disclosed revenue, named hyperscaler customers, valuation, or independently audited power savings. That leaves the commercial evidence where it belongs: ahead of PicoJool, inside qualification programs and production ramps.

Al Yuen's line in the announcement is the useful one: “Our focus now is execution.” At 200G per lane, PicoJool has earned a faster queue. The Series A decides whether the company can clear it with enough volume, reliability, and customer trust to keep the world's costliest processors busy.

BigHat Biosciences

A protein can bind the right target and still be a lousy drug.

It can aggregate, trigger the wrong biology, refuse to manufacture cleanly, or collapse under the first serious safety question. AI can propose a beautiful molecule. Patients, clinical teams, and chemistry eventually ask whether it behaves like medicine.

That gap is where BigHat Biosciences has built its business. The San Mateo company raised a $75M Series C co-led by DFJ Growth and Premji Invest - US, bringing company-reported funding to $223M.

The new capital lands after BigHat moved BHB810, its CDH17-directed antibody-drug conjugate, into a Phase 1/2 study for advanced gastric and gastroesophageal cancers. Its second program, BHB299, is a CEACAM6 T-cell engager in preclinical development, with human trials planned for 2027. This is capital arriving after the platform produced a clinical candidate, not another round attached to a software demo.

Peyton Greenside, CEO and co-founder, and co-founder Mark DePristo started BigHat in 2019 around a stubborn idea: better models need better experimental feedback. The company reports a roughly 1-week design-to-data loop, more than 2,000 molecules produced and characterized each week, and more than 20 assays measuring developability, affinity, and function.

The advantage is learning from the molecules that fail, measuring several drug properties on the same purified protein, and feeding that evidence into the next design cycle.

Premji Invest's thesis makes the wager clear. General-purpose models will keep improving and spreading. Proprietary data from hard molecular problems, especially negative data that rarely reaches public datasets, is harder to copy. So is the team required to move a design through manufacturing, an IND, and a first patient.

The $75M now has two jobs: expand the learning system and carry BHB810 and BHB299 toward clinical readouts. Those bets reinforce each other. Every program can generate better training data; every clinical milestone asks whether the platform's design choices survive a much less forgiving environment.

There is no human efficacy result to celebrate yet, and BigHat disclosed no valuation or investor check sizes. What exists is a cleaner line of sight between AI design, experimental evidence, and the clinic.

The next chapter will be written molecule by molecule, assay by assay, and patient by patient. BigHat now has the capital to keep that loop moving where the consequences become real.

Deeper Analysis 👇 https://devcuration.com/articles/bighat-biosciences-raises-75m-ai-designed-biologics

Ando

The human relay layer is becoming the strangest job in AI.

An agent writes the code, researches the market, or prepares the customer response. Then somebody copies the result into the channel, explains where it came from, adds the context the agent missed, and asks the team what should happen next.

We built software that can do the work, then kept it outside the room where work gets coordinated.

Sara Du started Ando after running into that gap while helping companies build MCP servers. Teams wanted agents inside Slack. The deeper problem was that workplace messaging still treated agents like apps even as companies expected them to act like teammates.

Ando has now emerged from stealth with $20M across pre-seed and seed financing. Accel led the pre-seed. Index Ventures and Emergence Capital led the seed. The exact split between the rounds, valuation, and ownership were not disclosed.

The product has channels, DMs, group conversations, and live conversations. Agents get identities, permissions, shared context, and a place in relevant work without waiting for a human to carry every output.

Ando is agent-agnostic. It says agents can connect work across channels, ask for missing information, and return judgment calls to people.

That promise carries a harder design problem than making a bot speak. A useful agent has to know when silence is more valuable, which context it may touch, whose instruction controls the work, and how the team can see what it did. Give every agent a voice without solving attention and governance, and collaboration software becomes a very expensive group chat nobody wants to open.

Ando reports users across 15 countries in software, real estate, finance, and professional services. Many teams are small. The figures are company-reported; revenue, retention, and independently measured productivity gains remain undisclosed.

The investor logic is legible. Small teams can change their communication habits quickly enough to show whether shared context compounds. If the pattern holds, Ando can move toward larger organizations that demand stronger administration, security, and migration proof.

Slack and Teams can add agents. Ando is making the more aggressive bet that agent coworkers change the architecture of messaging itself.

The $20M buys time to test that bet in the unglamorous moments: conflicting instructions, incomplete permissions, missing context, and the judgment call about whether the machine should speak before the next human message lands.

Deeper Analysis 👇 https://devcuration.com/articles/ando-raises-20m-agent-native-team-messaging

Parallax

A data center does not earn anything from a turbine promised for the 2030s.

That is the commercial premise behind Parallax. Founder and CEO Carl Schoeller has brought the company out of stealth with a $117M fundraise led by Founders Fund, Eclipse, Lux Capital, Diffusion, Greylock, and General Catalyst.

Parallax is building an approximately 10 MW gas turbine for onsite AI data-center power. The company is not trying to beat industrial turbines at the efficiency game they have spent decades perfecting. It is redesigning the machine around production speed.

A conventional turbine core can contain more than 2,000 parts. Parallax says 3-D printing can collapse that core into a handful. Lower combustion temperatures reduce fuel efficiency, but they also let the company use more common materials and pull from broader aerospace supply chains. The planned machine can travel in shipping containers instead of waiting for infrastructure designed around a different century of demand.

That trade-off is the story. AI developers can commit capital to compute on one schedule and discover that the grid, generation equipment, interconnection, and permits live on another. The IEA estimates data centers could consume about 950 TWh globally by 2030, with 15 GW to 27 GW of onsite natural-gas generation serving them by then, mostly in the United States.

Parallax wants to sell availability into that mismatch. The company is targeting a first prototype by the end of 2026, testing in 2027, and customer deliveries in 2028. Those dates are targets, not operating proof. No customer orders, installed machines, independently validated efficiency, reliability data, emissions profile, revenue, or valuation have been disclosed.

Schoeller previously co-founded Theseus, where a hackathon drone became a GPS-denied navigation company. A gas turbine is a far less forgiving object. Heat, rotating machinery, service intervals, permitting, and years of field performance do not care how persuasive the launch video was.

The investor syndicate is financing a disciplined industrial wager: give up some theoretical efficiency, simplify the hardware, widen the supply chain, and get power equipment to the customer while the compute still needs it.

$117M gives Parallax room to turn that wager into metal. The useful evidence will arrive at the test stand, then at the first data-center site, where delivery speed finally has to coexist with output, uptime, maintenance, and fuel cost.

Deeper Analysis 👇 https://devcuration.com/articles/parallax-raises-117m-ai-data-center-turbines