Make AI Agents Visible, Step by Step
AI agents can do remarkable work — but too often the same thing takes far too long, or has to be done several times before it finally works. The way out is to see the process itself: every step, and the time spent between them.
A Typical Problem With AI Agents
A typical problem I have with these agents is that either:
- a)the same thing takes a very long time to complete, or
- b)the same thing has to be done multiple times before it finally works.
One potential solution would be to visualize the process as a sequence of steps, while also capturing the time spent between each step. That would make it possible to see where the process is actually taking time, where work is being repeated, and where the biggest delays occur.
And that quality, in turn, significantly increases speed — even though it may theoretically be slower at the beginning.
I'm trying to get a sense of how large that difference in quality — and therefore the difference in speed — actually is, and how much more we can get done per week and per month.
— Ville, FollowLean
The Two Failure Modes
The same thing takes a very long time
The agent gets there in the end, but a task that should take minutes drags on. Without seeing each step, there is no way to tell which part of the run is actually eating the time.
The same thing has to be done multiple times before it finally works
The run fails, gets retried, fails again — and eventually succeeds on attempt four or five. Each retry costs time, and nobody can see which step keeps breaking.
A Potential Solution
Visualize the process as a sequence of steps — and capture the time spent between each one. Once the process is visible and timed, the problems stop being mysterious.
Visualize the process as a sequence of steps
Lay the agent's work out as an ordered process map instead of an opaque run log, so every step from start to finish is visible.
Capture the time spent between each step
Measure the duration of each step and each wait in between. The timing data turns a vague feeling of slowness into a specific, fixable list.
See where the process actually takes time
With steps and timings side by side, the slow segments stand out immediately — the step that always stalls, the handoff that always waits.
See where work is being repeated
Repeated steps and re-runs become visible as loops in the process, showing exactly where the agent does the same work more than once.
See where the biggest delays occur
Ranking steps by lost time shows where improvement pays off most, instead of guessing which part of the run to optimise first.
This Is Process Mapping
Steps, sequence, timing, repeats, bottlenecks — this is exactly what Lean process mapping does for human work. FollowLean brings the same visibility to any process, including the work agents do.
A Simple Example
Before
Write → 60 min/day
After
Write → 45 min/day
Improvement: −15 min/day — 15 minutes back, every single day, from one step alone.
Quality Significantly Increases Speed
It can feel slower at the beginning: mapping the process, timing each step, and measuring the repeats takes effort before anything visibly improves. That is the trade — a little structure now for a lot of speed later.
Once the process is visible, the wins compound: fewer repeated attempts, shorter waits between steps, and improvements aimed at the delays that actually cost the most time. Better process quality means the same agents simply get more done.
How Much More Can We Get Done?
The open question — and the one worth measuring — is how large the difference really is, week by week and month by month: