Kieran Clarke, home

01

Founder

Founder, agentic studio · Jun 2025 to present

One person, a studio of 60+ AI agents, and live products in fintech and consumer apps.

I founded and run it, directing agents across product, design, engineering, testing, infrastructure and operations. It takes products, projects and whole ventures from first design to live release, fast, under the controls that make agent work safe to ship.

How the studio is organised

It works like a small engineering firm where every role is an AI agent. I set the goals and the controls; every change is checked before it ships, and I give the final sign-off.

Org chart: Kieran, one person, at the top. Below him, the operator agent, which splits goals into tasks. Below it, 15 specialist AI agents in five groups: Build, Check, Research, Grow and Run. Under the chart, about 60 scheduled agents that run on their own.

Kieran, one personGoals, controls, final sign-off
Operator agentChief of staff: splits goals into tasks, assigns and chases them
  • Build

    • EngineerCode
    • FrontendScreens
    • InfraServers
    • OnchainBlockchain
  • Check

    • QATests
    • DesignerDesign review
  • Research

    • ResearcherSources
    • QuantMaths
    • DataDatasets
    • SignalTrends
  • Grow

    • ContentWriting
    • GrowthReach
    • CommunityPeople
  • Run

    • PMPlans
    • OpsUpkeep

About 60 scheduled agents

Monitors, health checks and daily briefings

16 roles plus about 60 scheduled agents

Goals go in at the top; work comes back checked, or it is sent back.

3,000+
tasks in
six weeks
Up to 10
specialists
at once
785
times work was sent
back and fixed

It is the rule I worked to in rail and at Amazon: nothing goes live until it has been checked.

An illustration of the studio’s delivery board around a real review finding. The issue Sign and submit orders is sent back at independent review with the real finding from 13 Oct: an order’s size was checked too early, so an out-of-date, too-small order could still have gone out; the illustration then shows it fixed, approved, signed off by a person and released.
A real finding from the review step (13 Oct, anonymised); the board is illustrative.

Controls, and what they caught

Real findings from the studio’s review step, October 2026, anonymised.

  • No AI agent approves its own work.
  • Every claim needs evidence.
  • I sign off every production database change.
  • Anything touching customer money gets two independent reviews.
  • Monitoring: Every share tap would have raised a false launch alert. Outcome: Sent back.
  • Payments: Overspending was logged but not stopped when new claims came in. Outcome: Stopped at review.
  • Security: Pressing set-up twice made a second login key and lost the first one. Outcome: Sent back.
  • Trading: An order’s size was checked too early, so an out-of-date, too-small order could still have gone out. Outcome: Sent back.
  • Trading: A missing or invalid region was let through instead of being stopped. Outcome: Sent back.

What the studio builds

Four rows from the app's live list of tokenised shares
ShareTodayPrice
NVDANVIDIA−4.26%$153.26
AAPLApple+0.85%$353.79
GOOGLAlphabet+3.86%$293.70
METAMeta Platforms+1.39%$625.76
Pictured: the consumer trading app, the only product shown. The studio designed, built and runs it. Four rows from its live list of tokenised shares, redrawn without logos.
What the studio builds across its products, in words, not pictured
CapabilityWhat it doesStatus
Real-time dataIndexers (services that read blockchain data), live market data and alerting.Around the clock
Full-stack productsWeb and mobile apps, interface to database.End to end
Financial flowsPayments and payouts on public blockchains.Contracts reviewed
Autonomous operationsAgents that watch systems, triage, and call a person.On call
Brand and mediaDesign systems, motion and 3D.Made in code

This site was built in a day by my agent studio, then refined over the following days. I signed off every change.

Work with me

How I help teams