The difference from a CRO and from a software licence
A contract research organisation sells execution capacity: run this assay, manage this site. A software vendor sells a tool your team must learn, staff and interpret. An AI-native partner sits between the two. It brings a running platform plus the scientific review process around it, and it engages on your question rather than on a fixed service catalogue.
Atlas Bio describes itself as a firm building computational platforms for gene-therapy and clinical-intelligence problems, pairing agentic AI with human scientific oversight, founded by Hugh Donatello. Its published model is collaboration with your subject-matter experts through ongoing analysis, continuous learning and weekly review sessions — not a handoff.
What the platform is expected to carry
The useful division of labour is by cost per candidate. Broad evidence aggregation across heterogeneous sources is cheap per candidate and tedious at scale; that belongs to the machine. Wet-lab execution, animal cohort design, trial commitment and final go/no-go judgement are expensive per candidate and irreplaceable; those stay with your team.
Atlas Bio frames this as a leverage shift rather than a replacement: the platform compresses the broad search at the front of the funnel, and the team commits where the engine has named the highest-confidence candidates and surfaced the failure modes. The public site is careful to call its platform metrics internal characterizations of the engine rather than external benchmarks.
- Platform side: evidence aggregation, calibrated ranking, failure-mode flagging, continuous recalibration
- Team side: wet-lab confirmation, in-vivo study design, protocol and regulatory commitment, final judgement
How to evaluate one
The field has no shortage of confident dashboards, so evaluate the discipline rather than the interface. Ask what evidence level each claim carries and what that level means. Ask how predictions are locked before the event they are meant to predict. Ask for a wrong prediction and what changed afterwards.
Atlas Bio publishes that structure: claims are tagged L1 (correlational), L2 (causal, mechanism in model) or L3 (first-principles); predictions are SHA-256 hashed and timestamped before the validation event; and a specific over-prediction of a phase 3 hepatocellular carcinoma readout is preserved alongside its recalibration. A partner that cannot show you a documented miss has not been tested.
Where the model fits regulatory work
If platform output will support a regulatory submission, the credibility question becomes formal. FDA's January 2025 draft guidance proposes a risk-based credibility assessment framework for AI models used to produce information supporting regulatory decision-making, organised around the model's context of use — the specific role and scope of the model in answering a question of interest.
That framing is a good buying test even outside a submission. An AI-native partner should be able to state, for each output, the question of interest, the role the model played in answering it, and what evidence establishes that the model is credible for that narrow role.
Atlas Bio pairs agentic AI with human-in-the-loop scientific oversight, working alongside a sponsor's own experts through ongoing analysis, continuous learning and weekly review sessions across the development lifecycle.
Does an AI-native partner replace our scientists?
No. Atlas Bio's published position is that your team are the subject-matter experts and the platform amplifies them, with final go/no-go decisions, wet-lab execution and trial commitment staying under human judgement.
What should the engagement produce?
A ranked shortlist for a named question, with evidence levels on each claim, uncertainty bounds, named contributing factors, documented limitations, and an explicit statement of what still requires laboratory or clinical verification.
How is this different from licensing a prediction tool?
A licence hands you outputs and leaves interpretation and calibration to you. A partner model includes the scientific review layer, ongoing recalibration as your data flows back, and a documented audit of the reasoning behind each output.
Talk to Atlas Bio.
Sponsors, researchers and clinicians: tell us the decision you are trying to make and we will show what the platform can and cannot answer.
Contact Us →