{ "format": "minglabs/v1", "surface": "insights-answer", "slug": "what-does-it-cost-to-run-an-ai-agent-in-production", "inLanguage": "en", "languages": [ { "language": "en", "label": "English", "path": "/insights/answers/what-does-it-cost-to-run-an-ai-agent-in-production", "url": "https://www.minglabs.com/insights/answers/what-does-it-cost-to-run-an-ai-agent-in-production", "jsonUrl": "https://www.minglabs.com/insights/answers/what-does-it-cost-to-run-an-ai-agent-in-production.json", "markdownUrl": "https://www.minglabs.com/insights/answers/what-does-it-cost-to-run-an-ai-agent-in-production.md" }, { "language": "de", "label": "Deutsch", "path": "/de/insights/answers/was-kostet-ein-ki-agent-im-laufenden-betrieb", "url": "https://www.minglabs.com/de/insights/answers/was-kostet-ein-ki-agent-im-laufenden-betrieb", "jsonUrl": "https://www.minglabs.com/de/insights/answers/was-kostet-ein-ki-agent-im-laufenden-betrieb.json", "markdownUrl": "https://www.minglabs.com/de/insights/answers/was-kostet-ein-ki-agent-im-laufenden-betrieb.md" } ], "kind": "answer", "pageType": "Answer", "question": "What does it cost to run an AI agent in production?", "anchorTermConceptSlug": "what-is-hybrid-organisation", "topic": "AI agent production costs and human supervision", "targetQueries": [ "What does it cost to run an AI agent in production?", "AI agent monthly operating costs including human review", "AI agent cost per completed task" ], "url": "https://www.minglabs.com/insights/answers/what-does-it-cost-to-run-an-ai-agent-in-production", "htmlUrl": "https://www.minglabs.com/insights/answers/what-does-it-cost-to-run-an-ai-agent-in-production", "partOf": "https://www.minglabs.com/insights/answers", "title": "What does it cost to run an AI agent in production?", "standout": null, "hook": "An AI agent’s running cost includes the software, the people who check its work and the people who keep it working. In our illustrative example, €320 in software becomes a €1,890 monthly budget once human time is included. That is €18.90 per task accepted for use, before setup and exceptional incidents.", "citableAnswer": "In MING Labs’ illustrative AI-agent cost model, 100 tasks accepted for use cost €1,890 a month: €320 for software, €1,250 for human review and €320 for ongoing ownership. All inputs are assumptions. Setup and exceptional incidents are excluded.", "hookSourceIds": [ "S1" ], "author": { "name": "MING Labs", "role": "Editorial team" }, "evidenceNote": "The worked example uses illustrative inputs. Sebastian Mueller’s operating figures are attributed to his published report; the underlying logs are not public.", "evidenceLabel": "Worked example", "summary": "The software bill is only part of an AI agent’s cost. Follow a MING Labs proposal-drafting example, then build a budget that includes the people around it.", "pillar": "Hybrid Organisation", "datePublished": "2026-10-01", "dateModified": "2026-10-01", "freshness": { "updated": "October 2026", "nextReview": "January 2027" }, "evidenceTier": "proprietary", "confidence": "B", "sources": [ { "id": "S1", "title": "AI agent production cost: illustrative inputs and calculation, version 1.0", "publisher": "MING Labs", "date": "2026-10-01", "url": "https://www.minglabs.com/insights/answers/ai-agent-cost-example.json", "supports": [ "€1,890 total, €18.90 per accepted task and €23.63 at 80 accepted tasks under disclosed assumptions", "cost-line decomposition and sensitivity calculation" ], "method": "Deterministic arithmetic on explicitly illustrative inputs; one monthly scenario and acceptance sensitivity. No measured production sample." }, { "id": "S2", "title": "How much of your week does an agent cost? Our number, and the Quarter Rule we hold it to.", "publisher": "Sebastian Mueller on LinkedIn", "date": "2026-09-23", "url": "https://www.linkedin.com/pulse/how-much-your-week-does-agent-cost-our-number-quarter-mueller-ymf8e/", "supports": [ "attributed report of €1.27 compute and 15–25 minutes review per proposal draft", "three cost lines; operator estimate; chosen Quarter Rule threshold" ], "method": "Public first-person operating report; underlying cost and review logs, sample size and proposal measurement dates are not public. Supports what was reported, not an independently verified benchmark." }, { "id": "S3", "title": "Building effective agents", "publisher": "Anthropic", "date": "2024-12-19", "url": "https://www.anthropic.com/engineering/building-effective-agents", "supports": [ "start with simpler architectures; evaluate added complexity against performance, cost and latency", "iteration limits and human checkpoints as design controls" ] }, { "id": "S4", "title": "About MING Labs: founders and operating model", "publisher": "MING Labs", "date": "Accessed 2026-10-01", "url": "https://www.minglabs.com/about", "supports": [ "Sebastian Mueller is a founding partner of MING Labs", "MING Labs builds and operates AI agents within its own team" ], "method": "Undated company page; the date records retrieval, not publication." } ], "faqs": [ { "q": "Is there a standard monthly price for an AI agent?", "a": "This example does not establish a standard price. Your budget depends on the workload, software charges, human review and ongoing ownership. Our €1,890 monthly example uses chosen assumptions; it is not a MING Labs quote and excludes setup and exceptional incidents." }, { "q": "Does the €1.27 proposal figure include human review?", "a": "No. Sebastian Mueller, a founding partner of MING Labs, reported €1.27 in computing costs per proposal draft, plus 15–25 minutes of human review. His September 23, 2026 article does not publish the underlying logs, sample size or measurement dates, so the figure should not be treated as a benchmark." }, { "q": "Should failed runs count towards the cost?", "a": "Yes. You still pay to produce and review work that gets rejected. Count the cost of all attempts, then divide by the results accepted for use. In our example, the same €1,890 budget costs €23.63 per accepted result when only 80 of 100 tasks are accepted. With none accepted, the cost per accepted result is undefined." }, { "q": "Does an hour saved mean an hour of salary saved?", "a": "No. A freed hour can help your team clear a backlog or spend more time with a client, but it does not automatically reduce payroll. Record what the team does with the time. Count a cash saving only when an expense is actually reduced or avoided." }, { "q": "What is the Quarter Rule?", "a": "Sebastian Mueller describes it as a rule MING chose: flag an agent for review when checking its work takes more than 25 percent of the hours it returns. It is a design threshold, not an industry standard or a research finding. The right limit depends on the work and the consequences of a mistake." } ], "parentArticleSlug": null, "relatedConceptSlugs": [ "what-is-hybrid-organisation", "what-is-the-abc-framework", "what-is-the-comprehension-obligation" ], "relatedArticleSlugs": [ "hire-dont-deploy" ], "metadata": { "title": "What does it cost to run an AI agent in production? — MING Labs", "description": "The software bill is only part of an AI agent’s cost. Follow a MING Labs proposal-drafting example, then build a budget that includes the people around it.", "canonical": "https://www.minglabs.com/insights/answers/what-does-it-cost-to-run-an-ai-agent-in-production", "inLanguage": "en", "openGraph": { "title": "What does it cost to run an AI agent in production?", "description": "The software bill is only part of an AI agent’s cost. Follow a MING Labs proposal-drafting example, then build a budget that includes the people around it.", "url": "https://www.minglabs.com/insights/answers/what-does-it-cost-to-run-an-ai-agent-in-production", "image": "https://www.minglabs.com/og/ai-agent-production-cost.png" }, "twitter": { "card": "summary_large_image", "title": "What does it cost to run an AI agent in production?", "description": "The software bill is only part of an AI agent’s cost. Follow a MING Labs proposal-drafting example, then build a budget that includes the people around it.", "image": "https://www.minglabs.com/og/ai-agent-production-cost.png" } }, "actions": [ { "id": "navigation", "label": "Get in touch", "accessibleName": "Get in touch", "kind": "navigate", "requiresBrowser": false, "url": "https://www.minglabs.com/contact" }, { "id": "discuss-agent-operating-cost", "label": "Talk to us about the first role you want to put through this exercise", "accessibleName": "Talk to us about the first role you want to put through this exercise", "kind": "navigate", "requiresBrowser": false, "url": "https://www.minglabs.com/contact?type=hybrid" }, { "id": "copy-briefing", "label": "Copy answer for my AI", "accessibleName": "Copy answer for my AI", "kind": "copy", "requiresBrowser": true, "contentUrl": "https://www.minglabs.com/insights/answers/what-does-it-cost-to-run-an-ai-agent-in-production.md" } ], "markdownUrl": "https://www.minglabs.com/insights/answers/what-does-it-cost-to-run-an-ai-agent-in-production.md", "contentMarkdown": "---\ntitle: \"What does it cost to run an AI agent in production? — MING Labs\"\ndescription: \"The software bill is only part of an AI agent’s cost. Follow a MING Labs proposal-drafting example, then build a budget that includes the people around it.\"\ncanonical: https://www.minglabs.com/insights/answers/what-does-it-cost-to-run-an-ai-agent-in-production\nlang: en\nlast-updated: 2026-10-01\n---\n\n# What does it cost to run an AI agent in production?\n\n[Home](/) [Insights](/insights) [Answers](/insights/answers) What does it cost to run an AI agent in production?\n\n[English](/insights/answers/what-does-it-cost-to-run-an-ai-agent-in-production)[Deutsch](/de/insights/answers/was-kostet-ein-ki-agent-im-laufenden-betrieb)\n\nAnswer\n\n\n\nAn AI agent’s running cost includes the software, the people who check its work and the people who keep it working. In our illustrative example, €320 in software becomes a €1,890 monthly budget once human time is included. That is €18.90 per task accepted for use, before setup and exceptional incidents.[\\[S1\\]](#source-s1)\n\nOne month of agent workIllustrative example · EUR\n\n**€1,890**100 accepted tasks \n€18.90 per accepted task\n\nRuntime**€320**\n\nReview**€1,250**\n\nOwnership**€320**\n\nIn this worked example, software costs €320; human review and ownership add €1,570. The inputs are illustrative. [See the assumptions and calculation](#worked-example).[\\[S1\\]](#source-s1)\n\nLast updated: October 2026 | Next review: January 2027 Worked example [Machine-readable record ](/insights/answers/what-does-it-cost-to-run-an-ai-agent-in-production.json)\n\nBy MING Labs · Editorial team\n\nThe answer at a glance\n\n1. [A €1.27 draft still needed 15–25 minutes of review.](#operating-evidence)\n2. [Put software, review and ownership in the same budget.](#cost-lines)\n3. [Divide that budget by the work your team can use.](#accepted-work)\n\n## 01 The €1.27 draft still needed a human\n\nSebastian Mueller is a founding partner of MING Labs, where we build and run AI agents alongside our human team.[\\[S4\\]](#source-s4) In his September 2026 operating report, he described a revealing pair of numbers: €1.27 in computing costs per proposal draft, plus 15–25 minutes of human review.[\\[S2\\]](#source-s2)\n\nA proposal draft is one step towards an offer you can send to a client. Someone still has to check whether it answers the brief and whether the company can stand behind it. The computing cost tells you what it took to generate the draft; the review time tells you part of what it took to make that draft useful.\n\nThat is where an agent budget should begin: follow the work until your team can use the result. The example above shows how the costs can add up. Below, we walk through the calculation so you can replace our assumptions with your own.\n\n## 02 The software bill is one of three costs\n\nSebastian separates running the system, reviewing its work and owning its operation.[\\[S2\\]](#source-s2) Each needs a line in the budget, because each pays for a different part of getting the work done.\n\n- **Runtime: the software bill.** Model calls, tools, hosting and other services the agent uses. Include retries as well as successful runs.\n- **Review: the time spent on its work.** Reading drafts, checking facts, making corrections and deciding what is ready to use.\n- **Ownership: the time spent keeping it useful.** Maintaining access, testing changes and adjusting the workflow when the job changes. Keep these hours separate from reviewing individual tasks.\n\nFor a worked example, imagine 100 proposal drafts in a month, all checked and accepted for use. Each gets 15 minutes of review. We have chosen the other rates below to make the calculation easy to follow; they are illustrative assumptions, not MING’s actual operating costs or a service quote.[\\[S1\\]](#source-s1)\n\nThe inputs behind our €1,890 monthly example \\[S1\\]\n| Cost line | Assumed inputs | Monthly cost |\n| --- | --- | --- |\n| Runtime | 100 attempts × €2 + €120 fixed | €320 |\n| Review | 100 attempts × 15 min ÷ 60 × €50/h | €1,250 |\n| Ownership | 4 hours × €80/h | €320 |\n| Total | 100 accepted tasks × €18.90 | €1,890 |\n\nIn this example, the software costs €320 and the people around it cost €1,570. Together, that is €1,890 a month, or €18.90 for each accepted draft.[\\[S1\\]](#source-s1) Changing the model price would affect only one part of that budget.\n\n## 03 A rejected draft still costs money\n\nA draft that gets rejected still takes resources to produce and time to check. Suppose the same 100 attempts yield only 80 results your team accepts, with no change in spending or review time. The cost per accepted result rises from €18.90 to €23.63.[\\[S1\\]](#source-s1)\n\nSame work and review, fewer accepted results \\[S1\\]\n| Accepted tasks | Monthly spend | Cost per accepted task |\n| --- | --- | --- |\n| 100 of 100 | €1,890 | €18.90 |\n| 80 of 100 | €1,890 | €23.63 |\n| 0 of 100 | €1,890 | No accepted outcome |\n\nDefine what “accepted” means before the trial. For a proposal draft, it could mean that the account lead has checked it and approved it for the intended next step. Counting drafts alone would miss that distinction. Our guide to [when to shut down an AI agent](/insights/answers/when-to-shut-down-an-ai-agent) looks at what happens when a team gets plenty of output but uses none of it.\n\n## 04 Measure a real workflow before you scale\n\nPick one recurring task and follow it from request to accepted result. Start with a week that includes ordinary work and awkward cases, then repeat if the workload varies. A few easy drafts will tell you less about the budget than a period with corrections, retries and handovers.\n\n1. **Record every attempt.** Keep the software costs for rejected work and retries in the total.\n2. **Time the human work.** Track review for each task and shared maintenance separately, so the same hour is counted once.\n3. **Mark what was accepted.** Use a named reviewer and the quality standard you would expect from a human doing the same task.\n4. **Compare equivalent work.** Work out the cost of completing the same job manually, then record what the team does with any time released.\n\nAn hour freed up can help someone clear a backlog or spend more time with a client. It does not automatically reduce the payroll. Show that extra capacity separately from expenses you actually reduce or avoid. Keep setup and integration costs visible too, even when you spread them across several months.\n\n## 05 If review eats the gains, revisit the role\n\nWhen checking an agent’s work takes almost as long as doing the job, examine the assignment. Is it too broad? Does the reviewer have a clear standard for accepting the result? Has the agent been given work that requires a person’s judgment?\n\nThe [ABC Framework](/insights/concepts/what-is-the-abc-framework) helps separate human judgment, shared expert work and routine operations. Use that distinction to narrow the role and make the handover clear. A [named human owner](/insights/answers/who-is-accountable-when-an-ai-agent-makes-a-mistake) still decides whether the result is good enough.\n\n> Budget for the people around the agent, then check whether the work is worth that time.\n\nA simpler technical approach can help too. Anthropic’s guidance on building agents recommends adding complexity when the performance gain justifies the cost and latency, with limits and human checkpoints where needed.[\\[S3\\]](#source-s3) The right change depends on what is taking the time: running the software, checking the output or maintaining the workflow.\n\n## 06 Start with one role and its full cost\n\nThe €1.27 proposal figure is useful once the human work sits beside it. For your own agent, start with one task, a clear definition of a usable result and a person who owns the outcome. That gives you a budget you can test before adding more agents.\n\nMING Labs’ [Hybrid Organisation work](/hybrid-organisation) helps teams design those roles and the systems around them. Bring a recurring task, its current workload and the time your team spends checking results. [Talk to us about the first role you want to put through this exercise](/contact?type=hybrid) .\n\nAbout the numbers\n\n- **The worked example:** one illustrative month, dated 1 October 2026, version 1.0. We assume 100 attempts, 15 minutes of review each, €2 runtime per attempt, €120 fixed runtime, €50 per review hour and four ownership hours at €80 per hour. The base case accepts all 100 tasks.[\\[S1\\]](#source-s1)\n- **What the example leaves out:** initial implementation, integration, tax, exceptional incidents and business losses from errors. Failed attempts stay in the cost total. With zero accepted results, cost per accepted result is undefined.\n- **Sebastian’s report:** he describes €1.27 compute and 15–25 minutes of review as measured, but does not publish the sample size, measurement dates or underlying logs. His four operator hours per agent per week are an estimate; the model above separately assumes four hours per month.[\\[S2\\]](#source-s2)\n- **Reproduce the example:** [open the inputs, formulas and results](/insights/answers/ai-agent-cost-example.json) , then replace the assumptions with your own observations.\n\nCitable short version\n\nIn MING Labs’ illustrative AI-agent cost model, 100 tasks accepted for use cost €1,890 a month: €320 for software, €1,250 for human review and €320 for ongoing ownership. All inputs are assumptions. Setup and exceptional incidents are excluded.\n\nThe worked example uses illustrative inputs. Sebastian Mueller’s operating figures are attributed to his published report; the underlying logs are not public.\n\n## Sources\n\n\\[S1\\]\n\n[AI agent production cost: illustrative inputs and calculation, version 1.0](/insights/answers/ai-agent-cost-example.json) MING Labs · 2026-10-01 Supports: €1,890 total, €18.90 per accepted task and €23.63 at 80 accepted tasks under disclosed assumptions, cost-line decomposition and sensitivity calculation Deterministic arithmetic on explicitly illustrative inputs; one monthly scenario and acceptance sensitivity. No measured production sample.\n\n\\[S2\\]\n\n[How much of your week does an agent cost? Our number, and the Quarter Rule we hold it to.](https://www.linkedin.com/pulse/how-much-your-week-does-agent-cost-our-number-quarter-mueller-ymf8e/) Sebastian Mueller on LinkedIn · 2026-09-23 Supports: attributed report of €1.27 compute and 15–25 minutes review per proposal draft, three cost lines; operator estimate; chosen Quarter Rule threshold Public first-person operating report; underlying cost and review logs, sample size and proposal measurement dates are not public. Supports what was reported, not an independently verified benchmark.\n\n\\[S3\\]\n\n[Building effective agents](https://www.anthropic.com/engineering/building-effective-agents) Anthropic · 2024-12-19 Supports: start with simpler architectures; evaluate added complexity against performance, cost and latency, iteration limits and human checkpoints as design controls\n\n\\[S4\\]\n\n[About MING Labs: founders and operating model](/about) MING Labs · Accessed 2026-10-01 Supports: Sebastian Mueller is a founding partner of MING Labs, MING Labs builds and operates AI agents within its own team Undated company page; the date records retrieval, not publication.\n\n## Frequently asked questions\n\nIs there a standard monthly price for an AI agent?\n\nThis example does not establish a standard price. Your budget depends on the workload, software charges, human review and ongoing ownership. Our €1,890 monthly example uses chosen assumptions; it is not a MING Labs quote and excludes setup and exceptional incidents.\n\nDoes the €1.27 proposal figure include human review?\n\nNo. Sebastian Mueller, a founding partner of MING Labs, reported €1.27 in computing costs per proposal draft, plus 15–25 minutes of human review. His September 23, 2026 article does not publish the underlying logs, sample size or measurement dates, so the figure should not be treated as a benchmark.\n\nShould failed runs count towards the cost?\n\nYes. You still pay to produce and review work that gets rejected. Count the cost of all attempts, then divide by the results accepted for use. In our example, the same €1,890 budget costs €23.63 per accepted result when only 80 of 100 tasks are accepted. With none accepted, the cost per accepted result is undefined.\n\nDoes an hour saved mean an hour of salary saved?\n\nNo. A freed hour can help your team clear a backlog or spend more time with a client, but it does not automatically reduce payroll. Record what the team does with the time. Count a cash saving only when an expense is actually reduced or avoided.\n\nWhat is the Quarter Rule?\n\nSebastian Mueller describes it as a rule MING chose: flag an agent for review when checking its work takes more than 25 percent of the hours it returns. It is a design threshold, not an industry standard or a research finding. The right limit depends on the work and the consequences of a mistake.\n\n## Quick Answers\n\n[What is a Hybrid Organisation?](/insights/concepts/what-is-hybrid-organisation) [What is the ABC Framework?](/insights/concepts/what-is-the-abc-framework) [What is the Comprehension Obligation?](/insights/concepts/what-is-the-comprehension-obligation)\n\n## More from Insights\n\n[Hybrid Organisation\n\n### You don't deploy an agent. You hire one.\n\nThe companies getting AI right stopped evaluating tools and started writing job descriptions. The shift sounds semantic. It isn't.\n\nRead article](/insights/articles/hire-dont-deploy)\n\n## Available actions\n\n- **Copy answer for my AI**: copies the [AI briefing](https://www.minglabs.com/insights/answers/what-does-it-cost-to-run-an-ai-agent-in-production.md) to the clipboard in the browser.\n\n---\n\nCanonical: https://www.minglabs.com/insights/answers/what-does-it-cost-to-run-an-ai-agent-in-production\nMachine-readable index: https://www.minglabs.com/llms.txt\n" }