Agents are rounded squares, people are circles, gates are lime. Each agent shows its accountable human, status and provenance.
Lines show how work travels. Select a node to see what came in, what went out and who touched it.
Human approvals sit where the risk is. A gate stays closed until its blocking flags are cleared. Real review requirements vary by material and market.7
Agentic products need components that interfaces built for screens never had. These twelve are the core set, and every one is used in the demo above.
Agents focus on their task and rarely come back to you. A coordinator checks that each has what it needs, raises a hand when something is wrong and keeps the process moving.
Set how much each agent may do: observe, flag, propose or act. Keep oversight where mistakes cost most.
Every error goes into a running log kept by an agent, so people decide what to clear first.
Every output says who made it and who approved it. With sources and version history, it shows how far to trust what you are reading.
Pick a contained process that will matter later. Map it, run it with real reviewers, learn from it, then carry the patterns that worked into the next one.
Every module and agent has an owner and a review-by date, so renewing or retiring it is routine.
Satisfaction scores reward tools people like. Agent Flow also tracks whether the work depends on them, and whether oversight is earning its place.
The share of users who would be “very disappointed” if the tool disappeared, the Sean Ellis question. A 40% benchmark is common, but it rests on one practitioner’s experience, so read it beside other measures.8
The share of outputs with a complete chain: author, approved sources, edits and sign-off.
How long a flag stays open.
How long work waits for a human, by gate.
Time from request to approved.
How far readers trust a piece of content, measured directly.
How often a person changes or sends back an agent’s output.
How often modules and agents are renewed or retired on schedule.
This page shows how Agent Flow thinks. What stands behind it is walked through live on a demo call.
Tell me a little about your team and what you want to see. I prepare each demo around your context, so I will reply within two business days to set a time. If you are looking to run a framework like this in your organization, say so in the message.
Prefer email? linton@lintonsmall.com
Twenty years of creative direction across broadcast, agency, and enterprise product design, including regulated industries. I build design systems and AI-enabled workflows that help teams move faster while keeping people accountable for every decision. Agent Flow is a framework in development, and this demo is a working prototype of it.
See my portfolioEach point in this page links to its source. Where a primary source was out of reach, the entry says why and points to the secondary reporting I used instead. The demo’s people, product and numbers are fictional.