--- title: "You don't deploy an agent. You hire one. — MING Labs" description: "When you hire a person, you give them a remit, the tools the role needs, a reporting line, and accountability for an outcome. The companies getting AI right d" canonical: https://www.minglabs.com/insights/articles/hire-dont-deploy lang: en last-updated: 2026-05-27 --- # You don't deploy an agent. You hire one. [Home](/) [Insights](/insights) [Articles](/insights/articles) You don't deploy an agent. You hire one. Hybrid Organisation The companies getting AI right stopped evaluating tools and started writing job descriptions. The shift sounds semantic. It isn't. Last updated: May 2026 | Next review: November 2026 Experiential [Machine-readable record ](/insights/articles/hire-dont-deploy.json) Sebastian Mueller · Founding Partner, MING Labs · May 2026 · 8 min read Tools vs Hires ## 01"Ask Martin" A few weeks ago a new colleague needed a feature shipped. She didn’t open a ticket or evaluate a tool. Someone told her: ask Martin. So she did, in a chat thread, the way you’d ask anyone. Martin built it and pushed it to production. Martin is not a person. He’s one of our agents. And the tell that this is working isn’t the code he shipped — it’s that a new joiner treated him like a colleague without being told to. For the first week or two, people sometimes can’t cleanly say which of their coworkers are human. That’s not a parlour trick. It’s the whole point. ## 02The wrong question, asked in every meeting Walk into most AI planning sessions and the question is “which tool should we buy.” Copilot or ChatGPT Enterprise. Seats, rollout, training. It’s the wrong question, and it quietly produces the wrong result. A tool is something a person operates. Buy more of them and you’ve handed already-stretched people more things to run. The work still sits on a human’s plate. Nothing structural moved. The question that moves things is the one you’d ask with headcount budget: what role do we need, and what should it own. ## 03What "hiring" actually changes When you hire a person you don’t hand them a capability. You hand them a remit, the tools the role needs, a reporting line, and accountability for an outcome. We build agents the same way — a job description goes in, an agent comes out, scoped to a role, and it takes a seat on the org chart.[\[S1\]](#source-s1) The shift is concrete, not semantic. “We need a foresight capability, let’s evaluate platforms” becomes “we’re opening a headcount for a foresight analyst” — filled by an agent. People stop asking which tool to open and start asking who to go to. The answer is a named colleague with a defined remit who happens to run on a model. Lola was hired, not deployed. Her remit is the pipeline, client intelligence, and the inbox. She doesn’t assist with the pipeline — she [owns](/insights/concepts/what-is-the-abc-framework) it. A deal slips, that’s hers. A stakeholder gets missed, that’s on her. People route those things to her the way they’d route them to a human chief of staff, and that routing is what a tool can never earn. ## 04The discipline is scoping, same as a hire The failure mode here is obvious the moment you frame it as hiring. You wouldn’t hire one person to run finance, IT, and marketing — they’d have no focus and do all three badly. People reach straight for that mistake with agents: one omni-assistant that does everything. It does everything thinly. One of our designers put it the way you’d put it to a hiring manager: give the agent one role with an adjacent set of skills, like a real teammate. Martin builds and ships product. Cody writes and deploys code. The night shift turns the backlog into reviewed drafts. Each owns a lane. The narrow remit isn’t a limitation — it’s what makes the agent accountable for an outcome, and accountability is the entire reason to do this instead of buying another tool. We learned the inverse the hard way: an agent we gave a vague “coordinate everything” brief produced nothing but noise, and we [shut it down](/insights/articles/we-fired-an-ai-agent) .[\[S2\]](#source-s2) ## 05Why this is harder than it sounds Writing a job description for an agent is harder than buying a licence, and that’s the point. You have to decide what the role owns, where the human line is, what “good” looks like, and who catches it when the agent gets something wrong — because it will. Lola isn’t flawless. Last month she over-escalated a client thread she’d misread. Fine. That’s a Tuesday. A colleague makes a judgement call, you catch it in the flow, the role survives the mistake. You only get that from something you’ve scoped as a [hire](/insights/concepts/what-is-hybrid-organisation) . You never get it from a tool.[\[S1\]](#source-s1) ## 06One thing to take from this > Stop shopping for tools. Start writing job descriptions. The unit of AI adoption isn’t software you roll out. It’s a role you fill. Decide what it owns, scope it like a hire, put it on the org chart, hold it to an outcome. Do that and the technology mostly takes care of itself. Skip it, keep evaluating tools, and you’ll keep wondering why the licences didn’t change anything. * * * MING Labs runs a hybrid organisation of humans and AI agents in production since January 2026. We help enterprises redesign work around human-agent teams. ## Sources \[S1\] Internal MING Labs operating model and agent org chart, January–May 2026MING Labs · 2026-05-27 Supports: Hire-not-deploy operating model, Named-agent remits and reporting lines, Scoping discipline, Lola Chief-of-Staff ownership pattern, Owned-mistake example \[S2\] Internal agent audit — decommissioned coordinator agentMING Labs · 2026-03-30 Supports: Vague-remit failure counter-example ## Frequently asked questions What does it mean to 'hire' an AI agent instead of deploying a tool? It means giving the agent a job description, a defined remit, the tools the role needs, and accountability for an outcome — then putting it on the org chart. A tool is operated by a person; a hired agent owns work itself, and colleagues route to it the way they'd route to a teammate. How broad should an agent's role be? Narrow. The same way you wouldn't hire one person to run finance, IT, and marketing, you shouldn't build one agent to do everything. A focused role with an adjacent skill set is what makes the agent accountable and reliable. What's the hardest part of treating an agent as a hire? Scoping and ownership, not the technology. You have to define what the role owns, where the human line sits, what good looks like, and who catches the agent's mistakes — because it will make them, exactly like a new hire. ## Quick Answers [What is a Hybrid Organisation?](/insights/concepts/what-is-hybrid-organisation)[What is the ABC Framework?](/insights/concepts/what-is-the-abc-framework) ## More from Insights [Hybrid Organisation ### We fired an AI agent after 13 days 764 messages. Not one that mattered. What we learned about the difference between tools and roles. Read article](/insights/articles/we-fired-an-ai-agent) [Hybrid Organisation ### We made our agents email people. That's when the AI started working. 95% of enterprise AI pilots show no return. The gap isn't the model. It's the last mile — and the last mile is boring, which is why almost everyone skips it. Read article](/insights/articles/the-last-mile) --- Canonical: https://www.minglabs.com/insights/articles/hire-dont-deploy Machine-readable index: https://www.minglabs.com/llms.txt