The Integrator seat is the single most affected role when an EOS® company brings AI in for real. The Visionary keeps casting the vision. The Implementer keeps coaching. The Integrator is the one whose day-to-day job description gets rewritten.
The short version: an Integrator used to harmonize the leadership team. Now the Integrator also runs the agent layer.
Gino Wickman's original definition of the Integrator in Traction® was the glue between the Visionary's ideas and the rest of the team. The Integrator removes obstacles, holds people accountable, runs the L10® meeting, owns the P&L, and translates the V/TO™ into execution. None of that changes. What changes is who reports up to the Integrator.
Before AI integration, the report-ups were all humans. After AI integration, several seats on the Accountability Chart are held by agents, and those agents report up to the Integrator the same way a human Director of Operations would. The Integrator is now accountable for the performance of a hybrid team.
What the new Integrator job actually contains
Five new responsibilities show up.
First, agent SOP ownership. Every agent seat needs a job description, a scorecard, and a written SOP. The SOP is the agent's actual operating instruction. The Integrator owns the quality of those SOPs the same way an Integrator already owns the quality of human onboarding. If an agent makes the same mistake twice and the SOP did not get updated in between, that is an Integrator failure, not an AI failure.
Second, scorecard fidelity. When humans owned the Scorecard, the failure mode was "I forgot to update the number." When agents own the Scorecard, the failure mode shifts to "the number is wrong because the upstream data source changed and the agent did not notice." The Integrator now has to audit agent data the way the Integrator used to audit human data.
Third, escalation triage. Agents escalate. A well-built agent never decides anything risky alone, it raises a flag to a human. The Integrator is the default escalation target unless the chart says otherwise. That means the Integrator's inbox now includes machine-generated flags that look like nothing the Integrator has seen before. Sorting real from noise is a new skill.
Fourth, model and platform decisions. Anthropic, OpenAI, and the rest will keep shipping new models. Claude Opus 4.7. Sonnet 4.6. ChatGPT new tiers every quarter. Each new model can break or improve agent behavior. The Integrator now owns a decision the Integrator never had to make: when do we upgrade, when do we hold, when do we switch vendors. This is a vendor management call, not a tech call, and it belongs in the Integrator seat.
Fifth, agent-to-agent conflict. Two human directors disagree, the Integrator mediates. Two AI agents start writing over each other's work or duplicating outreach, the Integrator now mediates that too. It feels strange the first time. It stops feeling strange about thirty days in.
What does not change
The Integrator's posture toward the Visionary does not change. The Visionary owns the V/TO™. The Integrator translates it. AI integration does not flip that direction. If anything, it amplifies it, because a fuzzy V/TO™ propagates into every agent's instructions and the Integrator is the one who has to chase down the fuzziness.
The Integrator's L10® meeting cadence does not change. Same Segue, Scorecard, Rock review, Customer/Employee Headlines, To-Do list, IDS, conclude. Same 90 minutes. The pre-meeting prep gets faster because agents pre-stage the data. The meeting itself stays disciplined.
The Integrator's accountability to the Visionary does not change. The Visionary still expects the Integrator to deliver the 1-Year Plan. The fact that some seats are held by AI does not let the Integrator off the hook for delivery. If anything it raises the bar, because agents do not get tired and do not take vacation.
How to staff this on day one
Most EOS® companies do not need to hire a new role. The Integrator absorbs the new responsibilities by reallocating time.
Time freed up because agents handle a lot of the Integrator's old reporting and follow-up work usually maps neatly to the time the Integrator now spends on agent governance. Net zero in the first 90 days. After that, most Integrators report getting time back, because the agents pre-stage so much of the L10® meeting and Scorecard.
If the company is larger, a Chief of Staff role can sit just under the Integrator and handle day-to-day agent management. At Sneeze It we use an agent named Radar in that Chief of Staff seat, which means the Chief of Staff function itself runs on AI, reporting up to the human Integrator.
What about Claude Code specifically?
For Integrators reading this and wondering where to start: Claude Code from Anthropic is the most direct on-ramp for building internal agents because it runs in the terminal where your operations live. SOPs become markdown files. Agents become slash commands. The whole stack stays inside a folder you can audit.
We run our entire agent army through Claude Code. Radar, Pepper, Dirk, Crystal, Dash, Pulse. Every one is a Claude Code agent with a job description in markdown and a scorecard the Integrator can read.
You do not have to use Claude Code. ChatGPT with Custom GPTs, OpenAI Assistants API, and a handful of other platforms can do this. The integration pattern is what matters, not the vendor.
FAQ
Does the Integrator need to be technical? No. The Integrator needs to be able to read a job description and a scorecard. The agent layer should be built so that non-technical Integrators can audit it. If the agent layer requires the Integrator to read code, the layer is built wrong.
Does the Visionary need to approve agent decisions? Only at the same level the Visionary approves any other big people decision. Hiring an agent for a key seat is a Visionary-Integrator conversation. Reviewing daily output is the Integrator's job.
Should the Integrator run the L10® meeting differently? No. Same agenda, same time box. The pre-meeting briefing gets richer because agents pre-stage it. The meeting itself stays the same.
EOS®, Entrepreneurial Operating System®, V/TO™, Level 10 Meeting®, L10®, Traction®, Rocks™, and Integrator are concepts and trademarks of EOS Worldwide, LLC. This article is an independent practitioner perspective and is not affiliated with or endorsed by EOS Worldwide.