Kaizen means change for the better. A Kaizen event, sometimes called a rapid improvement event, gathers the people who do a process for a few focused days to map it, find waste, try changes and put the working ones in place before they leave the room. Many organizations used the format for years in operations. It is just as useful now that AI agents are entering ordinary knowledge work, because the hardest questions are about the process, not the tool.
Why events suit AI-enabled work
When a team adds an agent to a process, three things tend to go wrong. The agent is bolted onto the old process, so it speeds up one step and creates a queue at the next. The handoffs between people and the agent are vague, so work stalls or is redone. And nobody has time to look at the whole flow, so problems stay invisible. A Kaizen event creates the time and the cross-functional view to fix those things together.
Before the event: scope and prepare
- Pick a specific process. Choose one with a clear start and end that people complain about or that an agent could plausibly help with. “Customer onboarding review” is better than “operations”.
- Write a short charter. State the problem, the boundaries of the process, what a better result would look like and what the team may and may not change.
- Name a sponsor, a facilitator and a team leader. The sponsor clears obstacles. The facilitator runs the room. The leader owns follow-up.
- Choose the team. Include the people who actually do each step, someone who receives the output, someone who understands the data and systems, and a person with a fresh view. Include whoever will own any agent.
- Gather baseline facts. How long does the work take, how often is it redone, where does it wait? Rough honest numbers from the team are enough. Do not invent them.
During the event: a four-day shape
Events usually run two to five days depending on the process and who can be released. This shape works for most.
- Day 1, see the current state. Brief the team on waste and flow. Map the process step by step on a wall: who does what, what the handoffs are, where work waits or loops back. Mark steps that are routine and rule-based, steps that need judgment and steps where information is lost.
- Day 2, find causes and options. Ask why at each problem step until you reach a cause you can change. Sort the steps into menial and meaningful. Brainstorm changes, including where an agent could take a routine step, and where a person must stay.
- Day 3, design and test. Draw the future state with explicit handoffs, approval points and a fallback if the agent is wrong. Try the changes on live or realistic work immediately, in small trials, and record what happened.
- Day 4, standardize and report. Keep what worked. Write the new standard work, including agent instructions and review points. Assign every remaining action an owner and a date. Present the results to the sponsor.
Design questions specific to agents
- What exactly does the agent receive, and what does it hand on? Write it as a contract.
- Where does a person check the work, and what are they checking for?
- How will we know the agent is wrong or drifting? Who is told?
- What knowledge does the agent rely on, and who keeps it current?
- What do we do on the day the agent is unavailable?
- Which tasks did we remove from people, and what meaningful work will they spend that time on?
After the event: make it stick
Most gains are lost in the weeks after the event, when daily work resumes. Hold a short weekly check on the action list. Review at thirty, sixty and ninety days: is the new process being followed, is the agent behaving as designed, has anything drifted? Capture lessons from the event in your shared library so the next team starts ahead. This is how a single event turns into scaled learning.
Pitfalls to avoid
- Choosing a process too large to improve in a week.
- Letting managers design the process while the people who do it watch.
- Automating a broken step instead of removing it.
- Skipping the baseline, then being unable to tell whether anything improved.
- No follow-up. An event without follow-up is a workshop.
If you are not sure which process to pick, start with a value stream identification session. Our book 100 AI-Agile Teams offers patterns for how people and agents can share work, and our facilitators can run an event with your team.