Early-stage teams lose hours every week to repetitive work. We diagnose the workflow slowing your team down most, build the agent that takes it off their plate, and stay until the evals pass and your team adopts it.
// install prompt
Paste this into your agent and let it handle the rest. It interviews you to find the highest-leverage workflows, and writes a report explaining what to build and why.
Install the AI Workflow Audit for this agent. The AI Workflow Audit is a free agent that interviews a founder about one messy, repetitive workflow, then produces a build-ready AI Workflow Audit Report: a current-state map, the bottlenecks, an AI opportunity scorecard, a recommended first workflow, a V1 system design, human-review points, risks, and an eval plan. It understands the workflow before it recommends anything. It is a directory of 13 files, not a single file: SKILL.md plus its templates, rubrics, and examples. SKILL.md links to those by relative path, so install the whole directory or it will run shallow. If you are Claude Code, install it cleanly with the one command below. It fetches every file into ~/.claude/skills/ai-workflow-audit/ and is safe to re-run: curl -fsSL https://raw.githubusercontent.com/ryan-hennebry/ai-operator/main/scripts/install-claude-skill.sh -o /tmp/install-audit.sh && bash /tmp/install-audit.sh remote If you cannot run shell commands (Codex, Cursor, or a plain chat), do not fake it. Read the agent directly from https://raw.githubusercontent.com/ryan-hennebry/ai-operator/main/skills/ai-workflow-audit/SKILL.md , follow every relative link it references from the same directory, and run the audit from those instructions. If you cannot fetch URLs or write files at all, stop and tell me to switch to Claude Code, Codex, or Cursor. Before you start, ask me exactly one question: Which workflow should we audit first? 1. Founder-led GTM and market intelligence (the usual starting point) 2. Customer support or operations 3. Hiring and recruiting ops 4. Something else, I'll describe it Then begin stage 1. Interview me about my company and that workflow before proposing anything. Do not name a tool, model, or architecture until the workflow is mapped. Finish by producing the full AI Workflow Audit Report and telling me what to build first and why.
// the method
A first version, with review and evals built in.
Put it in front of a real user.
Prove it earns its place.
Stay until the team uses it.
Keep it working and improving.
The audit is free and yours to keep. The paid build sprint takes the workflow it recommends and builds it.
// ai opportunity scorecard
Inside the audit, each repetitive workflow is scored on the same dimensions, so the first thing we build is the one with the clearest payoff.
Competitor sites, pricing, and hiring are public; the weekly manual check is pure repetition; change detection plus a short brief is easy to verify.
Research the account before the call, capture the notes after. One trigger, one owner, one surface, the CRM. A human checks the summary before it lands.
Mentions across LinkedIn, Reddit, X, and forums are public and constant. The agent flags what matters so you reply in time; a human still filters signal from noise.
Turn one strong asset into posts, clips, and snippets from templates. The source is yours, so quality is high, but a human edit gate is non-negotiable on tone.
Pull the GTM numbers and write the weekly read against target. A model strength, held at Strong until the inputs are clean and reachable.
More buyers ask an LLM who does X before they reach your site. Worth tracking how you are cited now to get a baseline; build it properly once it is sending real pipeline.
Illustrative example of how the audit scores founder-led GTM workflows. The readings are assessment judgements (AI fit, time cost, data readiness), not measured client results.
// the deliverable
Fourteen sections, in order, from company context to the suggested next step.
AI Workflow Audit Report 14 sections build-ready hands off to the build sprint
// who it is for
Not sure which you are? The audit will tell you, for free.
// the offer
Run it in your own coding agent or chat. It works through five stages and produces a build-ready report. No call, no cost, and no guessing: the audit shows whether there is a workflow worth building.
One workflow, about a month. Scoped, built, deployed, and evaluated with human review and logging from day one. Picks up where your report leaves off and runs to a live workflow the team uses.
A deployed workflow drifts. We watch failures and iterate the system, so it keeps earning its place.
Watch from £1,000/mo
Evals re-run on every change, drift and failure monitoring, regression fixes, and a monthly adoption report.
Iterate £1,500/mo
Everything in Watch, plus active iteration and capacity to build your next workflow.
// questions
No. The audit is the only way in. You run it yourself inside your own coding agent, and you qualify yourself so we both know whether there is anything worth building.
The audit is free. If it finds a workflow worth building, the build sprint starts from £4,500, and an improvement retainer starts from £1,000 a month. You only pay for what earns its place.
Claude Code is the full experience. Codex, Cursor, and plain chat are honest fallbacks. It is a directory of files, so install the whole thing, not a single file.
A build-ready AI Workflow Audit Report: fourteen sections, from company context to the recommended first workflow, a V1 system design, human review points, risks, and an eval plan. You keep it whether we work together or not.