AI Project Management for R&D Programs in Biotech and Biomedicine

A research program is a moving set of questions, experiments, papers and people. Most of the pain in running one is not any single task. It is that the plan goes stale, the evidence lives in five places, and nobody has a current picture of what depends on what. An AI project manager for R&D is the idea of handing the bookkeeping of that picture to software, so that scientists and program leads spend their time on the science and the decisions.

This post sets out what that can mean in practice for a biotech or biomedical team, and where to be sceptical.

Why R&D programs are hard to keep on track

• Plans decay. A plan written at kickoff stops matching reality within weeks. Updating it is nobody's favourite job, so it doesn't happen.

• Dependencies are invisible. Experiment C waits on a result from B, which waits on a reagent. Without the dependency chain written down, delays surprise people.

• Evidence is scattered. Papers in one folder, data in another, code in a repository, discussion in Slack.

• Status reporting eats time. Writing the weekly update for stakeholders is work that produces no science.

What an AI program manager can reasonably do

Be specific about the claim. The useful version is not an autonomous manager. It is an assistant that keeps the structure current and reports on it. In codon.bio, that takes the following forms.

A staged research plan with dependencies and a critical path. The plan breaks a program into stages, records what depends on what and shows the critical path, the chain of tasks that sets the earliest finish. Progress is tracked from the project's own conversations: as work happens in the project's chats, tasks move from planned to active to done, each with a note on why and links to the chats that show the work. Statuses you set by hand are left alone. You still edit the plan's structure yourself, or regenerate it, but you no longer have to remember to tick things off.

One workspace for the whole program. A project holds the papers, data and code together, with connections to Google Drive, Dropbox, GitHub and Slack. It can also search public GitHub code, so a question about how a method is usually implemented is answered from real repositories. Questions asked of the project draw on all of it.

Proactive insights. The system does not wait to be queried. Each week a project's insights page summarises what is new since the previous week, for example recent papers that bear on the project's question, with citations and a note on what the new evidence does and does not settle. A useful paper or a suggested task can be added to the project in one click.

Weekly research reports. A report is synthesised across your active projects and sent by email, with an executive summary, what changed since last week and open questions, so the weekly update starts as a draft you edit instead of a blank page.

Team structure. Teams are invitation-only, with admin, manager and member roles and a shared workspace. A project can belong to a team or be shared with individual collaborators, and chats and files can be shared with one or more teams.

Where the science and the management meet

What distinguishes this from a generic task tracker is that the same system can answer scientific questions. When the plan says "confirm genetic support for target X", the project can run that question against GWAS, gnomAD, ClinVar, OpenTargets and the literature, with every claim cited. The plan then picks up that work: the step moves from planned to done, linked to the chat that holds the sourced answer.

A way to start small

1. Create one project for one live question, not the whole portfolio.

2. Load what you already have: the key papers, the dataset, the repository.

3. Ask for a plan, then edit it. Treat the first version as a draft to argue with.

4. Run a real step through it, such as an evidence-gathering question, and check that the plan picks it up.

5. Read the first weekly report and judge whether it saves you the hour it should.

Where to be sceptical

• It does not know what you have not told it. Wet-lab status, reagent delays and informal agreements only appear if someone records them. A plan is only as current as its inputs.

• It does not set priorities for you. The critical path shows what constrains the schedule. Deciding what matters is a leadership call.

• Treat generated reports as drafts. Read them before sending them to anyone.

Questions people ask

What is an AI project manager?

Software that maintains a project's plan and status from its inputs and reports on it, so people don't have to do the bookkeeping. It assists a human program lead; it does not replace one.

Is it different from tools like a task tracker?

A task tracker holds tasks you enter. Here the plan is connected to the project's papers, data and conversations, the same workspace answers scientific questions with citations, and task progress is tracked from that work.

Does it work for teams?

Yes. Teams have roles and a shared workspace. A project can belong to a team, and chats and files can be shared with one or more teams.

What does it cost?

There is a free Individual plan. Pro is $100 per month and Lab is $120 per seat per month, with Enterprise on request. With a .edu email, Pro is $50 per month and Lab is $60 per seat per month. See the pricing page.

Try it on one live program

Open a project in codon.bio for the question you are working on this month and see whether its plan and weekly report match your own picture.

Related: AI drug discovery for biotech and AI literature review.