The slowest part of inventing a new medicine is not the chemistry. It is the assembling. A drug that clears the lab still has to become a data package - thousands of pages of clinical evidence, trial design, statistical output and regulatory prose, cross-checked until a reviewer at the FDA can sign it. That job routinely takes months, and it runs on software written before most founders were born. Enjamb, a company in Y Combinator's Spring 2026 batch, is betting the whole assembly line can be handed to AI agents.
The pitch is unusually literal. Enjamb says its agents run a drug program end to end, "from preclinical research through clinical trials and approval." Not a copilot bolted onto one step, but a workspace that carries the work across every hand-off - the place where evidence gets synthesized, a trial gets designed, the statistics get programmed, and the submission gets written.
The name is a tell. Enjambment is the poetry term for a sentence that spills past the line break without stopping. A drug program is one long sentence broken across a dozen departments, and most of the delay lives in the breaks.
The ProblemPharma software doesn't talk to agents
Walk into a large drug company and you find the same stack everywhere: Benchling for lab data, Veeva for documents, Medidata for trials, SAS for statistics. These systems are trusted, validated, and closed. There is no clean way to point a modern AI agent at them and let it do real work. Enjamb's first move was to build that connection layer, so agents can act on top of the tools teams already use - without a migration, and without asking a Fortune-100 pharma to rip anything out.
That "no migration" detail is the strategy, not a footnote. Enterprise pharma will not swap a validated system on a startup's say-so. Every one of those systems went through years of validation to satisfy regulators, and pulling one out means re-validating everything downstream of it. By layering on top instead of replacing, Enjamb turns the incumbents' lock-in into its own on-ramp: the harder a system is to remove, the more valuable it is to have an agent that can finally work with it.
The gap Enjamb is filling has a name inside the industry: there is no standard way to connect an autonomous agent to these closed platforms. General AI tools can summarize a paper, but they cannot reach into a validated trial database, run the analysis the way the protocol requires, and return an artifact a regulator will accept. Bridging that last part - real execution inside the systems of record, not just conversation about them - is the piece Enjamb claims to have built first.
How Enjamb positions itself: a thin agent layer over software pharma already trusts.
The ProductA browser tab that does the paperwork
Open Enjamb and you get a working environment, not a chat box: editors for Word, Excel and PowerPoint, plus sandboxed Python and R for running analysis. Inside it, agents execute the scientific and regulatory tasks a program demands - reading the literature, drafting methodology, producing datasets, and stitching together the submission. The company's headline claim is speed with a paper trail: FDA submission packages in around 48 hours rather than roughly eight months.
Figures are the company's own; independent verification pending. Bar lengths are illustrative.
Speed alone would be a liability here. In regulated work, a single wrong number can sink a trial, so Enjamb's second claim matters as much as its first: it reports datasets with a 98% first-pass FDA compliance rate and roughly 23x fewer errors than a general model like GPT-4.5. Whether or not those exact figures hold up, the emphasis is the point - in this market, accuracy is the product.
The design choice that ties it together is provenance. Enjamb keeps every source, file, tool call, code run and document change attached to the output it produced. A reviewer can trace a line in the final submission back to the paper, the dataset, and the exact step that generated it. The agents do the assembling; a human still does the signing.
Practically, that changes what a team can attempt. A literature review that would occupy a scientist for weeks can be run against a very large corpus of papers and returned in hours, with citations mapped back to full text. A methodology can be drafted with the vendor and procurement detail already filled in. Datasets can be analyzed in a sandboxed environment and turned into publication-quality figures with the statistics run and shown. A manuscript or a grant proposal can be drafted against the right template. None of these are new tasks - they are the tasks that already eat the calendar. Enjamb's argument is that agents can do the first ninety percent and leave the judgment to the people.
"Enjamb adapts to your team's protocols, documents, and institutional knowledge - and automates the hardest parts behind the scenes."Enjamb, on how the platform fits existing teams
The Customers500 pharma staff, fast
Adoption arrived quicker than the funding. Enjamb says that within weeks of launch, more than 500 employees at large pharma companies - Johnson & Johnson, AbbVie, Bristol Myers Squibb, Merck and Sanofi - were using the platform. Those are logos most enterprise startups spend years chasing, showing up before a Series A. The buyers are the people inside a program: clinical, biostatistics and regulatory teams who feel the eight months most.
It is worth reading that number for what it is - early usage, not signed enterprise contracts. But usage is how a wedge becomes a platform, and Enjamb's framing has widened accordingly. The company started closer to "Cursor for life-science R&D," a tool for labs and researchers, and grew into running the full program as the customers moved up-market. Watching the positioning change in public is instructive: the workspace was the wedge, and the program is the market.
The competitive picture depends on where you stand. Against the contract research organizations that pharma has long paid to do this work, Enjamb is faster and cheaper, but unproven at scale. Against the incumbent software - Benchling, Veeva, Medidata, SAS - it is a companion, not a rival, at least for now. Against general-purpose AI that teams already paste sensitive work into off the books, Enjamb's answer is that a governed, auditable, domain-built system beats a clever chatbot the moment a regulator asks how a number was produced.
The FoundersYoung, and research-first
Enjamb was founded in 2025 by Rayan Mubarak, the CEO, and Maadhav Deekshitha, the CTO. Mubarak comes from the science he is now automating: a first-author cancer machine-learning paper published in JMIR, reported to beat the prior state of the art by 20%, with work presented at scientific toxicology and computational-biology conferences. He is also, reportedly, 19. Deekshitha's background is systems and hardware AI - work at Dell's AI Lab and Broadcom's R&D on chip-level optimization.
That pairing - a researcher who has felt the paperwork and an engineer who has shipped low-level performance work - reads as deliberate. One knows what a compliant dataset has to look like; the other knows how to make agents reliable enough to produce one.
In a market where a wrong number ends a trial, the founding bet is that agents can be accurate enough to be trusted with the last mile - the submission.
The Money & The Market$650K into a $140B habit
Enjamb raised a $650K pre-seed, announced in April 2026, backed by Y Combinator and Founders Inc. It is a small round for a large target: much of drug development's grunt work is currently outsourced to contract research organizations and armies of specialists, a market measured in the tens of billions. Enjamb's alternative is not another point tool but an execution layer that could absorb work now split between CROs, consultants, and general-purpose AI that teams use off-label.
The risks are the obvious ones for anyone building here. Regulated buyers move slowly and validate everything; the accuracy and speed claims will meet scrutiny before they meet contracts; and the incumbents whose systems Enjamb rides on could decide to build their own layer. The counter is that Enjamb is small on purpose - two founders serving enterprise pharma is only possible because the agents do the assembling - and that it chose the one problem where being trustworthy, auditable and boring is a feature.
The business model follows from all of this. Enjamb sells software to the teams inside a program rather than billing by the project the way a contract shop does. That means its value compounds as agents take on more of the work and as the provenance trail becomes something reviewers rely on. A CRO gets paid more when work takes longer; Enjamb gets paid more when a customer trusts it with the next step. Those incentives point in opposite directions, and in a field that has grown used to the first, the second is the interesting part.
There is also a quieter observation buried in the founders' backgrounds. The people most frustrated by pharma's paperwork tend to be the scientists who trained to do research and instead spend their days formatting datasets and chasing document versions. Enjamb is being built by someone who was recently one of them. That is not a guarantee of success - plenty of insider tools never leave the lab - but it is the kind of specific, felt problem that tends to produce a product people actually keep open.
For now, the company is doing the unglamorous thing well: treating drug development as the document-and-evidence problem it largely is, and trying to carry the sentence all the way to the period.