Work / Projects

Building at the intersection of context-aware AI, human purpose, and future-of-work systems.

These projects are connected by one question: what would software look like if it helped people hold more context, make better decisions, and delegate work without losing authorship?

Primary company / Contexta

Accountable context for agents doing consequential work.

The wedge is not a better chat window. It is the moment a team needs to trust an agent across systems, over time, and still recover the evidence behind its next move.

Bring one workflow

Buyer: Product and operations teams putting agents into recurring, consequential workflows.
Broken workflow: Context is scattered across meetings, documents, and tools, so nobody can inspect why an agent believes or acts.
Product now: An inspectable layer for memory, source provenance, unresolved assumptions, and decision history before action.
Next milestone: A design-partner workflow carried from briefing to action with its evidence and human checkpoints intact.

01 · Primary company · Context-aware AI infrastructure

Contexta

A system for giving AI agents durable working memory, provenance, and context that can be inspected, updated, and trusted.

Context should be treated as a first-class product surface: versioned, attributable, recoverable, and useful across the whole lifecycle of work.

Who it is for: Product and operations teams delegating recurring, high-context work to AI agents.
What exists now: The inspectable memory and provenance layer is in active product development.
Next proof: Take one real workflow from briefing to action while a human can inspect what the agent knew and why it acted.

Designing memory structures that separate facts, preferences, decisions, tasks, and unresolved assumptions.

Exploring interfaces that let people see why an agent believes something before they let it act.

Building toward agent workflows where context survives across meetings, documents, tools, and time.

Bring a design-partner workflow

02 · Exploring · AI-native workflow and virtual employees

Amino.ai

An exploration of virtual employees: agents that can own bounded work, report progress, ask for judgment, and improve through feedback.

The future of work is not just automation. It is a new organizational interface where software can hold responsibility without erasing human purpose.

Who it is for: Teams testing where an AI collaborator can own work without hiding uncertainty.
What exists now: Workflow primitives under test: delegation, review, escalation, feedback, and memory.
Next proof: Carry one bounded role from a clear brief through review, escalation, and accountable completion.

Mapping what a virtual employee must know before it can responsibly perform a role.

Testing workflow primitives for delegation, review, escalation, and memory.

Studying how teams can gain leverage without turning every craft into a queue of automated tasks.

Ask about Amino.ai

03 · Researching · Purpose, careers, and future-of-work systems

Work for Future

A research and product thread about meaningful work, career agency, and the tools people need to shape better futures.

If AI changes the price of execution, the scarce resource becomes direction: what people choose to build, why it matters, and how work remains humane.

Who it is for: People who want career decisions to account for purpose, fit, and long-term agency.
What exists now: A research and product thesis connecting meaningful work, career context, and human authorship.
Next proof: Turn purpose matching from a broad ideal into one testable decision surface.

Writing about the tension between automation and purpose.

Collecting patterns for career systems that help people compound taste, skill, and agency.

Looking for product shapes that make long-term contribution easier to see and coordinate.

Collaborate

04 · Shipped · Open-source developer infrastructure

Presetter

A configuration system for TypeScript projects and monorepos that keeps shared engineering decisions composed, reviewable, and current.

The point is not fewer config files. It is fewer unexplained decisions.

Who it is for: TypeScript teams maintaining consistent decisions across projects, packages, and time.
What exists now: A public CLI, scoped preset ecosystem, v9 release, migration guide, and production release pipeline.
Next proof: Make each upgrade read like a reviewed decision instead of another copied configuration file.

Version 9 gives the preset ecosystem one public namespace and a stricter modern baseline.

Runtime, Storybook, and monorepo presets turn repeated setup into explicit composition.

Provenance-backed publishing and resumable releases treat the delivery system as part of the product.

Explore Presetter Explore the repository Read the v9 field note

05 · Publishing · Essays, diagrams, and prototypes

Field Notes

A public notebook for mathematics, complexity science, design, architecture, and the emotional texture of building.

Serious technical work benefits from a more humane public surface: precise enough to be useful and personal enough to be remembered.

Who it is for: Builders who want the engineering argument and the human stakes in the same notebook.
What exists now: Seven public essays spanning agent reliability, context systems, product craft, and open source.
Next proof: Keep translating the thesis into code, diagrams, and claims that can be tested in public.

Short essays on agent reliability, context systems, and product taste.

Visual diagrams for ideas that are easier to understand spatially than verbally.

Small prototypes that test whether an idea changes how people think or work.

Read essays

Come build this

A small, founder-led product problem—still close to code, users, and first principles.

I want to meet product-minded engineers who care about agent reliability, knowledge systems, and interfaces that make trust inspectable. The immediate problem is to take one high-context workflow from memory capture through accountable action.

Based in London. Start with the problem; then decide what the right collaboration should become.

Talk about building Contexta

How we work

01Write decisions down while the context is still warm.
02Expose uncertainty early; quiet ambiguity becomes expensive later.
03Ship narrow vertical slices that a real person can challenge.
04Keep human judgment visible at every delegation boundary.

What I want the work to prove.

AI agents become useful when they understand the context of the work, not just the syntax of the request.

Production trust requires provenance: people should be able to inspect sources, assumptions, decisions, and actions.

The future of work should increase human agency. Automation is valuable when it gives people more room for judgment, craft, and direction.