LatchBio

LatchBio is a San Francisco-based company building data, compute and AI infrastructure for biotechnology R&D. Its original platform is a browser-based, no-code environment in which researchers can import files from existing data stacks and run bioinformatics pipelines and visualization tools — including RNA-seq analysis, CRISPResso2 and AlphaFold — without provisioning servers or writing code. The platform is organized around several components: Latch Data (storage in a private workspace backed by Amazon S3, with instrument and cloud-source connectors for Benchling, AWS, Dropbox, GCP and Azure, audit trails and fine-grained permissions), Latch Registry (a spreadsheet-like relational database that links samples, metadata and results, with a Python CRUD API), Latch Workflow Manager (workflow orchestration supporting Nextflow natively and a Python SDK that auto-generates user interfaces from typed function parameters), and Pods (hosted notebooks and dashboard apps such as RShiny, Dash and Streamlit, with access to configurable CPU, memory and GPU resources). The company states compliance with SOC-2 Type II and HIPAA frameworks.

More recently the company has repositioned around research and deployment of AI agents for biology. Under a research arm branded benchmarks.bio, it publishes open, reproducible benchmarks measuring how frontier AI agents perform on real-world biological data across spatial biology, single-cell, epigenomics, preclinical pharmacology and biosecurity; published or preprinted benchmarks include SpatialBench, SpatialBench-Long, scBench, scBench-Long, EpiBench, TxBench-PP, BioSecBench-Refusal and BioSecBench-Surveillance. Its deployment arm, agent.bio, is a public interactive sandbox where agents analyze user-supplied data in formats such as FastQ, CSV, Excel, H5AD and TIFF.

The agent products are packaged for two audiences: solution providers (kit, assay, instrument and service vendors), who can offer white-labeled analysis packages that customers redeem alongside a purchased kit, and biopharma or R&D teams seeking unified workflows, faster collaboration and traceability across the drug discovery process. The company says its tooling accelerates analysis for more than 40 kits, instruments or services and is aimed at bench scientists, bioinformaticians and computational biologists.

Founding story

The three co-founders — Alfredo Andere (CEO), Kenny Workman (CTO) and Kyle Giffin (COO) — met at UC Berkeley. Workman studied bioengineering and computer science while working as a machine learning and software engineer at Berkeley Lab, the Joint BioEnergy Institute, Asimov and Serotiny, where he was repeatedly unable to find off-the-shelf vendors to store files, analyze data or set up machine learning workflows. Giffin studied cognitive neuroscience and data science and worked as a data analyst in consumer products and investment banking, where data analysis was comparatively streamlined; Andere studied electrical engineering and computer science and interned at Facebook and Google, contrasting Google's advertising data pipelines with the infrastructure he saw at biotech companies. After roughly a year and a half working in the space and after helping about half a dozen biotech companies set up bioinformatics infrastructure, they concluded a generalizable platform was needed, conducted several hundred user interviews, and dropped out of Berkeley with one semester remaining to build the company.

Business model

LatchBio sells software to biotechnology and pharmaceutical organizations. Its platform is offered to companies on a paid basis while being free for academic users. Agent products are packaged two ways: a white-labeled analysis experience sold to kit, assay and service providers to accompany their products, and a data and compute agent sold to internal biopharma R&D teams. Sales are conducted through demo requests and direct calls with the research and engineering team.

Paid commercial contracts with biotech and pharma customers, with the platform free for academic researchers. The company expected roughly $1 million in ARR by the end of 2022.

Traction

By March 2023 the company reported over 30 paying customers running more than 10,000 workflows per month, including the Innovative Genomics Institute, bit.bio and Eligo Bioscience. It expected about $1 million in ARR by the end of 2022. Its agent sandbox is described as supporting more than 40 solution providers and more than 300 biopharmas and R&D labs worldwide, with analysis coverage for over 40 kits, instruments or services.

Latest developments

The company now presents itself as researching and deploying AI agents for biology through two programs: benchmarks.bio, publishing verifiable agent benchmarks across spatial, single-cell, epigenomics, preclinical pharmacology and biosecurity domains; and agent.bio, a public interactive agent sandbox. Its most recent listed preprint, BioSecBench-Surveillance, built in collaboration with Aclid, comprises 100 evaluations of whether AI agents can infer the correct analysis pipeline from raw sequencing data and sparse surveillance context across seven categories from taxonomic classification to genetic-engineering detection; across 3,962 gradable attempts from 16 model–harness pairs, the top configurations (Claude Opus 4.8 with Pi, and GPT-5.5 with Codex) scored about 50.2%.