R / 01Experiments
CLE
Learning to coordinate.
Experiments in how AI agents coordinate, keep useful state between tasks, and repair that state when evidence changes.
Research focus: coordination, persistent state and repair in bounded tests.
Business / AI / independent research
Make AI useful
in your business.
I’m Tommy Talbot, an entrepreneur working across business, AI and independent research. Only Then Labs brings that work together: practical workflows, clear decisions, and evidence that stays with the work.
Practical work.Clear decisions.Lasting context.
01 / Work with meOne workflow. A clear next step.
AI Workflow
Review.
Discuss a workflow A focused assessment of one recurring business workflow. We map the people, tools, decisions and handoffs, identify bottlenecks and risks, and define what is worth testing next.
What you receive
- Current workflow map
- Prioritized improvements
- A proposed pilot with measurable acceptance criteria
- Clear human review and approval points
Scope, fee, access and timing are agreed before work begins. Implementation is scoped separately.
02 / About Tommy
Business experience.
A research mindset.
My background spans service businesses, operations, sales and customer relationships. My current work explores how people and AI agents coordinate, preserve useful context and carry evidence through changing workflows.
03 / Independent researchCoordination / continuity / evidence
Intelligence
should compound.
Useful context should carry forward as people, agents and tools change. These research strands explore how to preserve that context, check what supports it, and make the next handoff more reliable.
R / 02Prototypes & experiments
WITNESS / ProofRoute
Evidence across a handoff.
WITNESS tests whether a handoff matches a signed record. ProofRoute explores signed receipts for explainable infrastructure recommendations.
Separate prototypes. Signed records check integrity; evidence and permission still require review.
R / 03Design concept
ROOT
Continuity with clear limits.
Design work on what an agent may inherit, who can authorize its next action, and how permission changes carry forward.
A separate family-recovery concept explores organizing records and access instructions. It is distinct from the agent-authority work.
R / 04Experimental prototype
MIRROR
Know when evidence is enough.
Controlled tests of how systems respond when evidence changes, including when they should withhold a conclusion.
Synthetic evaluation of when to withhold a conclusion. Real-world validation remains open.
Start a conversation
Have a workflow
in mind?
Tell me what you’re trying to accomplish
and where the work gets stuck.