The posting
AI Biologist - Biologics Development
Latch builds rigorous scientific benchmarks (benchmarks.bio http://benchmarks.bio) for AI agents across biology. We work with frontier labs and pharma to measure whether agents can handle real biological workflows: the messy, multi-layer reasoning that matters in the field.
We’re seeking an AI Biologist to join our TxBench team, pushing what AI can achieve in therapeutic antibody development.
About the Role. You will be the scientific authority on ground truth, testing whether agents can reason like an expert biologics development scientist. The benchmark covers monoclonal antibodies, Fc-fusions, and bispecifics in the window after lead nomination and before IND. AI agents today struggle to interrogate raw analytical traces and preclinical data, separate real quality-attribute failures from analytical artifacts, and make advance, engineer, or kill calls an expert would stand behind. You will work with software engineers and scientists to build the tasks and datasets behind our benchmarks. Your judgment will set the standard agent outputs are held to, including:
Build the task
- Turn real development workflows into file-based tasks with structured outputs and deterministic grading: developability triage, analytical/CMC characterization, PK and FcRn reasoning, immunogenicity, effector function and safety, and first-in-human dose projection
- Write gold-standard answers from raw data, with tolerances that reflect real replicate, integration, and fitting variability
- Curate data fixtures: developability panels, glycan LC-MS, charge-variant CE, intact/subunit MS, SEC-MALS, FcRn/FcγR sensorgrams, concentration-time PK, and immunogenicity and effector assays
- Partner with software engineers to turn your judgment into new workflows
Set the hard calls
- Test whether agents can tell a real liability from an artifact, such as a co-eluting glycan peak, a misassigned charge variant, or a single assay flag not confirmed by orthogonal readouts
- Test whether agents match effector function and affinity to the mechanism, rather than assuming “more is better”
- Test whether agents distinguish linear from target-mediated clearance, and judge a molecule against its target product profile
Judge the agent
- Interrogate agent reasoning step by step, and challenge conclusions that sound right but wouldn’t survive a development review or go/no-go decision
- Pinpoint where agents go wrong: why a step fails, what an expert would do instead, and what evidence would change the call
- Score agent confidence, rewarding explicit uncertainty over overconfident calls
Requirements
- PhD in protein engineering, biochemistry, analytical chemistry, pharmacology, or a related field
- Hands-on analysis of antibody development data. You’ve done the analysis yourself, not just supervised it
- Deep expertise in at least one area (analytical/CMC characterization, large-molecule DMPK, protein engineering and developability, or non-clinical safety) and working fluency across the others
- Command of how readouts map to critical quality attributes, PK and effector properties, and release or comparability decisions
- Python or R for data analysis, including parsing chromatograms and electropherograms, kinetic fitting, or PK modeling
- AI-ready from day one: daily use of Claude, Cursor, or similar agents, with the judgment to spot where model reasoning is technically correct but scientifically unsound
- Confidence to overrule an AI agent when the evidence demands it, and the openness to update when shown wrong
- Ability to calibrate confidence and communicate uncertainty rather than defaulting to binary calls
- Bonus: antibody format and Fc engineering experience (IgG subclasses, YTE/LS, effector-silencing, bispecific architectures)
- Bonus: familiarity with MAM, ICH Q6B/Q5E, FDA immunogenicity guidance, IMGT, SAbDab, and IEDB
Culture. We practice radical transparency: you’ll know how the company is doing, what’s working, and what isn’t. It’ll always be us against the problem, never each other. We’re opinionated but respectful, and we bring solutions with our problems. You decide how you work; we judge the output. We use AI constantly, so what we reward is original expert judgment, not AI output passed off as your own. You’ll start on the bench that fits your expertise and can move across teams as you build a track record. People who do well here and bring leadership experience can move quickly into senior roles. Outside of work, we have company outings most months, team off-sites, a Latch run club, cycling, boxing, ice skating, and soccer. Feel free to add to the list.
Compensation & Logistics
- Contract: 1099 (or W-8BEN internationally), no end date
- Pay: guaranteed for weeks 1-4 while you ramp, then paid per certified task. Most people in this position earn $120K–$180K+ (uncapped upside); certify twice the tasks/week, earn twice the pay that week.
- Commitment: full-time, on your own schedule. Most of the team works 40+ hours a week.
- Meetings: a few pod check-ins a week. The rest of your time is yours.
- Location: remote (anywhere), hybrid, or onsite in SF. We prefer onsite.
- Onsite perks: catered lunch and dinner daily, a waterfront office in Mission Bay, and monthly parking when available
- Work authorization: US citizens, permanent residents, and OPT holders. We can’t support STEM OPT extensions or sponsorship.
Interview Process
We move fast: about 2–3 weeks from your first call, to offer.
1. Intro Screen - Latch Talent Team
2. Take-Home Project - HackerRank
3. Technical Interview - Member of Technical Staff
4. Culture Interview - C-Suite
5. Offer
