{ "format": "minglabs/v1", "surface": "insights-namespace-answers", "id": "https://www.minglabs.com/insights/answers", "url": "https://www.minglabs.com/insights/answers", "htmlUrl": "https://www.minglabs.com/insights/answers", "name": "MING Labs Insights — Answers", "description": "Question-framed pages answering buyer questions about agents in production, from MING Labs' operating evidence. Built only when the query passes the Hyperize SUCHE qualification gate.", "inLanguage": "en", "dateModified": "2026-07-17", "partOf": "https://www.minglabs.com/insights", "publisher": { "name": "MING Labs", "url": "https://www.minglabs.com", "sameAs": [ "https://www.linkedin.com/company/ming-labs" ] }, "count": 6, "items": [ { "slug": "should-ai-agents-hold-their-own-credentials", "url": "https://www.minglabs.com/insights/answers/should-ai-agents-hold-their-own-credentials", "jsonUrl": "https://www.minglabs.com/insights/answers/should-ai-agents-hold-their-own-credentials.json", "title": "Should AI agents hold their own credentials?", "question": "Should AI agents hold their own credentials?", "targetQueries": [ "should AI agents hold their own credentials", "can AI agents hold API keys", "AI agent credential management", "how do AI agents authenticate securely", "AI agent secrets management", "AI agent security architecture" ], "standout": "MING Labs' agents hold no credentials at all. A gateway holds every secret, and a rogue agent can do nothing the gateway has not authorised.", "hook": "No. In MING Labs' production fleet, agents hold no credentials at all: no OAuth tokens, no API keys, no database credentials. A gateway layer holds every secret; agents request actions, the gateway authorises and executes them against a whitelisted tool surface declared in each agent's job description. The consequence is architectural, not procedural: an agent that goes rogue, or gets prompt-injected, can do nothing the gateway has not authorised. OWASP names excessive permissions as a root cause of agentic risk; this is the corresponding bright line.", "citableAnswer": "According to MING Labs' security architecture, AI agents hold no credentials: no OAuth tokens, no API keys, no database credentials. A gateway layer holds all secrets and executes whitelisted actions on the agent's behalf, so a compromised agent cannot reach anything the gateway does not authorise.", "summary": "AI agents should not hold their own credentials. In MING Labs' fleet, all secrets live in a gateway layer that authorises and executes actions against a whitelisted, versioned tool surface declared per agent. A prompt-injected or rogue agent can change what it asks for, not what it is granted. OWASP's LLM06:2025 names excessive permissions as a root cause of agentic risk; keeping agents credential-free is that finding turned into architecture.", "pillar": "Hybrid Organisation", "evidenceTier": "proprietary", "confidence": "B", "datePublished": "2026-07-17", "dateModified": "2026-07-17", "freshness": { "updated": "July 2026", "nextReview": "January 2027" }, "parentArticleSlug": null, "relatedConceptSlugs": [ "what-is-hybrid-organisation" ], "relatedArticleSlugs": [ "hire-dont-deploy" ], "sources": [ { "id": "S1", "title": "MING Labs operating record: fleet security architecture, gateway-mediated authentication, and per-agent tool surfaces", "publisher": "MING Labs (internal)", "date": "2026-07-01", "supports": [ "no agent holds OAuth tokens, API keys, or database credentials; all outbound authentication is mediated by a gateway", "each agent runs isolated in its own container, with no shared filesystem and no shared network namespace", "each agent's tool surface is declared in its job description, versioned, and whitelisted; anything not declared is denied", "encryption keys are separated per region and rotated quarterly" ] }, { "id": "S2", "title": "LLM06:2025 Excessive Agency, OWASP Top 10 for LLM Applications 2025", "publisher": "OWASP GenAI Security Project", "date": "2024-11-18", "url": "https://genai.owasp.org/llmrisk/llm062025-excessive-agency/", "supports": [ "excessive functionality, excessive permissions, and excessive autonomy named as the root causes of excessive agency", "recommended mitigations include restricting extensions to minimum necessary permissions and requiring human approval for high-impact actions" ] } ] }, { "slug": "what-belongs-in-an-ai-agents-audit-trail", "url": "https://www.minglabs.com/insights/answers/what-belongs-in-an-ai-agents-audit-trail", "jsonUrl": "https://www.minglabs.com/insights/answers/what-belongs-in-an-ai-agents-audit-trail.json", "title": "What belongs in an AI agent's audit trail?", "question": "What belongs in an AI agent's audit trail?", "targetQueries": [ "what belongs in an AI agent audit trail", "AI agent audit trail requirements", "AI agent logging requirements", "how to audit AI agents", "EU AI Act logging requirements for AI agents", "AI agent traceability" ], "standout": "Regulators ask one question: show me exactly what the agent did. MING Labs traces ten fields per agent action to answer it inside 24 hours.", "hook": "Ten fields per action. MING Labs' standard traces, for every agent action: agent identity and version, the triggering event, the job-description scope and authority level in force at execution, tool calls with parameters, data accessed, the output artefact and its destination, cost and duration, human review, and the downstream consumption chain. The test is the regulator's question: show me exactly what this agent did, when, with what inputs, who authorised it, and what the outputs were used for, answered inside 24 hours.", "citableAnswer": "According to MING Labs' operating standard, an AI agent's audit trail records ten fields per action, from agent identity and version through triggering event, authority level, tool calls, data accessed, output artefact, cost, and human review, to the downstream consumption chain, so that 'show me exactly what this agent did' is answerable within 24 hours.", "summary": "An AI agent's audit trail should reconstruct any action from ten fields: agent identity and version, triggering event, job-description scope at execution, authority level, tool calls, data accessed, output artefact, cost and duration, human review, and downstream consumption. The organising principle: every entry maps to a named element of the agent's job description, so the trail is the job description in motion, not a stream of opaque events. The EU AI Act makes automatic event logging a legal requirement for high-risk systems from August 2026.", "pillar": "Hybrid Organisation", "evidenceTier": "proprietary", "confidence": "B", "datePublished": "2026-07-17", "dateModified": "2026-07-17", "freshness": { "updated": "July 2026", "nextReview": "January 2027" }, "parentArticleSlug": null, "relatedConceptSlugs": [ "what-is-the-comprehension-obligation", "what-is-hybrid-organisation" ], "relatedArticleSlugs": [ "we-fired-an-ai-agent" ], "sources": [ { "id": "S1", "title": "MING Labs operating record: per-action trace standard, retention policy, and fleet review cadence", "publisher": "MING Labs (internal)", "date": "2026-07-01", "supports": [ "ten fields traced per agent action, each mapped to a named element of the agent's job description", "100 percent of fleet tasks logged with timestamp, cost, and outcome; no off-ledger work", "retention: 90 days operational, 12 months for security events, contract plus two years for client work", "review cadence: weekly automated audits across every agent, monthly manual review for high-severity work" ] }, { "id": "S2", "title": "Regulation (EU) 2024/1689 (EU AI Act), Article 12: record-keeping", "publisher": "European Union", "date": "2024-07-12", "url": "https://artificialintelligenceact.eu/article/12/", "supports": [ "high-risk AI systems shall technically allow for the automatic recording of events (logs) over the lifetime of the system", "the requirement applies from 2 August 2026" ] }, { "id": "S3", "title": "Regulation (EU) 2024/1689 (EU AI Act), Article 26: obligations of deployers of high-risk AI systems", "publisher": "European Union", "date": "2024-07-12", "url": "https://artificialintelligenceact.eu/article/26/", "supports": [ "deployers must keep the logs automatically generated by a high-risk AI system for at least six months" ] } ] }, { "slug": "who-is-accountable-when-an-ai-agent-makes-a-mistake", "url": "https://www.minglabs.com/insights/answers/who-is-accountable-when-an-ai-agent-makes-a-mistake", "jsonUrl": "https://www.minglabs.com/insights/answers/who-is-accountable-when-an-ai-agent-makes-a-mistake.json", "title": "Who is accountable when an AI agent makes a mistake?", "question": "Who is accountable when an AI agent makes a mistake?", "targetQueries": [ "who is accountable when an AI agent makes a mistake", "who is responsible for AI agent errors", "AI agent accountability", "can an AI agent be held accountable", "RACI for AI agents", "who is liable for an AI agent's output" ], "standout": "MING Labs runs its agent fleet on one RACI rule: an agent can be Responsible for work, but Accountability never leaves a named human.", "hook": "The named human who owns the outcome. That answer is the design, not an accident: in MING Labs' fleet, an agent can be Responsible for a piece of work, but it is never Accountable. Accountability stays with the human whose name is on the firm, the same place it sat before the agent was hired. The EU AI Act points the same way: human oversight must sit with named natural persons who have the competence, training and authority to exercise it.", "citableAnswer": "According to MING Labs' operating rule, an AI agent can be Responsible for a work unit but never Accountable: when an agent makes a mistake, accountability sits with the named human owner of the outcome, the same place it sat before the agent was hired.", "summary": "In MING Labs' operating model, agents are R, never A. An agent can own a work unit operationally and be Responsible for it; organisational accountability stays with a named human. Every autonomy grant is a logged event with a reason and an approver, so the accountability question has a written answer before the mistake happens.", "pillar": "Hybrid Organisation", "evidenceTier": "proprietary", "confidence": "B", "datePublished": "2026-07-17", "dateModified": "2026-07-17", "freshness": { "updated": "July 2026", "nextReview": "January 2027" }, "parentArticleSlug": null, "relatedConceptSlugs": [ "what-is-the-comprehension-obligation", "what-is-hybrid-organisation" ], "relatedArticleSlugs": [ "we-fired-an-ai-agent" ], "sources": [ { "id": "S1", "title": "MING Labs operating record: fleet decision rights, authority log, and RACI per workflow", "publisher": "MING Labs (internal)", "date": "2026-07-01", "supports": [ "agents are Responsible, never Accountable; accountability stays with a named human", "proposal workflow RACI: agent drafts as Responsible, a senior partner is Accountable for the outcome", "every change to an agent's authority is a logged event with timestamp, approver, and reason", "two incidents per quarter revert an agent to the previous autonomy level" ] }, { "id": "S2", "title": "Regulation (EU) 2024/1689 (EU AI Act), Article 26: obligations of deployers of high-risk AI systems", "publisher": "European Union", "date": "2024-07-12", "url": "https://artificialintelligenceact.eu/article/26/", "supports": [ "deployers shall assign human oversight to natural persons who have the necessary competence, training and authority", "deployers must keep automatically generated logs for at least six months" ] } ] }, { "slug": "how-often-do-ai-agents-get-shut-down", "url": "https://www.minglabs.com/insights/answers/how-often-do-ai-agents-get-shut-down", "jsonUrl": "https://www.minglabs.com/insights/answers/how-often-do-ai-agents-get-shut-down.json", "title": "How often do AI agents get shut down?", "question": "How often do AI agents get shut down?", "targetQueries": [ "how often do AI agents get shut down", "AI agent decommissioning rate", "how many AI agent pilots fail", "how long do AI agents last in production", "what share of AI agents reach production" ], "standout": "Most agent programs count launches. MING Labs counts retirements: 15 to 20 percent of its agents are decommissioned in their first quarter.", "hook": "Expect to shut down 15 to 20 percent of your AI agents within their first quarter. That is MING Labs' own decommissioning rate across a production fleet running since January 2026, with three retirements by name: Major Tom, CHRISTINE, and SM3CB v1. The rate is not a defect of the program; it is what an honest shutdown criterion produces. Industry projections point the same way: Gartner expects over 40 percent of agentic AI projects to be canceled by the end of 2027.", "citableAnswer": "According to MING Labs' operating data, 15 to 20 percent of the AI agents it puts into production are decommissioned within their first quarter: three retirements by name (Major Tom, CHRISTINE, SM3CB v1) in a fleet running since January 2026.", "summary": "MING Labs decommissions 15 to 20 percent of its AI agents within their first quarter. That rate is the output of an honest shutdown criterion, not a sign of failure. Gartner projects over 40 percent of agentic AI projects will be canceled by the end of 2027; a fleet where nothing is ever retired is unmeasured, not stable.", "pillar": "Hybrid Organisation", "evidenceTier": "proprietary", "confidence": "B", "datePublished": "2026-07-16", "dateModified": "2026-07-16", "freshness": { "updated": "July 2026", "nextReview": "January 2027" }, "parentArticleSlug": "we-fired-an-ai-agent", "relatedConceptSlugs": [ "what-is-the-comprehension-obligation", "what-is-hybrid-organisation" ], "relatedArticleSlugs": [ "we-fired-an-ai-agent" ], "sources": [ { "id": "S1", "title": "MING Labs operating record: fleet registry, retirements, and decommissioning rate", "publisher": "MING Labs (internal)", "date": "2026-07-01", "supports": [ "15 to 20 percent of agents decommissioned within their first quarter", "three named retirements: Major Tom, CHRISTINE, SM3CB v1", "fleet live in production under named human owners, since January 2026" ] }, { "id": "S2", "title": "We fired an AI agent after 13 days (Major Tom message audit, n=764)", "publisher": "MING Labs", "date": "2026-03-30", "url": "https://www.minglabs.com/insights/articles/we-fired-an-ai-agent", "supports": [ "Major Tom shut down after 13 days and 764 messages with zero acted-on output" ] }, { "id": "S3", "title": "Gartner Predicts Over 40% of Agentic AI Projects Will Be Canceled by End of 2027", "publisher": "Gartner", "date": "2025-06-25", "url": "https://www.gartner.com/en/newsroom/press-releases/2025-06-25-gartner-predicts-over-40-percent-of-agentic-ai-projects-will-be-canceled-by-end-of-2027", "supports": [ "over 40 percent of agentic AI projects projected to be canceled by end of 2027", "cancellation drivers named by Gartner: escalating costs, unclear business value, and inadequate risk controls" ] }, { "id": "S4", "title": "State of Agentic Orchestration and Automation 2026 (Coleman Parkes survey)", "publisher": "Camunda", "date": "2026-01-14", "url": "https://camunda.com/press_release/three-quarters-of-organizations-admit-gap-between-agentic-ai-vision-and-reality/", "supports": [ "only 11 percent of agentic AI use cases reached production in the past year", "71 percent of organisations say they use AI agents" ], "method": "Online survey by Coleman Parkes of senior IT and business decision makers and enterprise software architects at organisations with 1,000+ employees, fielded 23 September to 23 October 2025", "n": 1150 } ] }, { "slug": "when-to-shut-down-an-ai-agent", "url": "https://www.minglabs.com/insights/answers/when-to-shut-down-an-ai-agent", "jsonUrl": "https://www.minglabs.com/insights/answers/when-to-shut-down-an-ai-agent.json", "title": "When should you shut down an AI agent?", "question": "When should you shut down an AI agent?", "definedTerm": "Acted-On Rate", "targetQueries": [ "should you shut down an AI agent", "how do you know an AI agent is not adding value", "AI agent produces output nobody uses", "when to kill an AI agent pilot" ], "standout": "Week-one polish is noise. The Acted-On Rate, the share of agent output a human acts on, is the shutdown signal.", "hook": "Shut down an AI agent when its Acted-On Rate stays at zero: nobody uses or acts on its output, however busy it looks. Volume is not the signal; use is. MING Labs shut down its first agent, Major Tom, after 13 days and 764 messages because not one output changed a decision. A slow colleague improves that rate week over week. Noise does not.", "citableAnswer": "MING Labs shut down its first AI agent, Major Tom, after 13 days and 764 messages because not one output was ever used or acted on: capability without accountability, an Acted-On Rate of zero.", "summary": "The shutdown signal for an AI agent is a flat Acted-On Rate: output volume without a single human acting on it. MING Labs shut down its first agent after 13 days and 764 unused messages, and runs its fleet on the same measure.", "pillar": "Hybrid Organisation", "evidenceTier": "proprietary", "confidence": "B", "datePublished": "2026-07-05", "dateModified": "2026-07-16", "freshness": { "updated": "July 2026", "nextReview": "January 2027" }, "parentArticleSlug": "we-fired-an-ai-agent", "relatedConceptSlugs": [ "what-is-the-comprehension-obligation", "what-is-the-abc-framework" ], "relatedArticleSlugs": [ "agents-grading-themselves" ], "sources": [ { "id": "S1", "title": "We fired an AI agent after 13 days (Major Tom message audit, n=764)", "publisher": "MING Labs", "date": "2026-03-30", "url": "https://www.minglabs.com/insights/articles/we-fired-an-ai-agent", "supports": [ "Major Tom shut down after 13 days and 764 messages", "zero acted-on output as the shutdown reason", "capability without accountability diagnosis" ] }, { "id": "S2", "title": "MING Labs operating record: fleet task reviews and autonomy levels", "publisher": "MING Labs (internal)", "date": "2026-07-01", "supports": [ "Acted-On Rate as the fleet's standing shutdown and promotion signal", "healthy agents show acted-on output within the first two weeks", "autonomy levels promoted and demoted under named human ownership", "15 to 20 percent of agents decommissioned within their first quarter" ] } ] }, { "slug": "can-your-team-still-do-the-work-your-agents-do", "url": "https://www.minglabs.com/insights/answers/can-your-team-still-do-the-work-your-agents-do", "jsonUrl": "https://www.minglabs.com/insights/answers/can-your-team-still-do-the-work-your-agents-do.json", "title": "How do you prove your team can still do the work your agents do?", "question": "How do you prove your team can still do the work your agents do?", "anchorTermConceptSlug": "what-is-the-comprehension-obligation", "targetQueries": [ "how to prove staff can still do the work AI does", "how to demonstrate AI literacy of employees", "what does EU AI Act Article 4 actually require you to do", "how to stop losing skills when AI agents take over work", "how to test whether a human can still do an AI agent's job" ], "standout": "The law demands AI literacy and names no test. MING's test: reproduce the agent's work unaided, in 30 minutes, every quarter.", "hook": "You prove it by testing it, on a schedule, with a consequence attached. MING Labs writes a Comprehension Obligation into an agent's job description: the operator must reproduce that agent's three primary outputs unaided, within 30 minutes, rehearsed each quarter. Fail the rehearsal and the agent drops one autonomy level until the operator is current again. It is a test, not a training record. The EU has required AI literacy since February 2025 without ever saying how to demonstrate it.", "citableAnswer": "MING Labs writes a Comprehension Obligation into an agent's job description: the human operator must be able to reproduce the agent's three primary outputs unaided within 30 minutes, rehearsed quarterly, and a failed rehearsal demotes the agent one autonomy level until the operator is current again.", "summary": "The way to prove a team can still do the work its AI agents do is to make them do it, on a schedule, with something at stake. MING Labs' Comprehension Obligation requires the operator to reproduce an agent's primary outputs unaided within 30 minutes each quarter; failing costs the agent an autonomy level.", "pillar": "Hybrid Organisation", "evidenceTier": "proprietary", "confidence": "B", "datePublished": "2026-07-14", "dateModified": "2026-07-14", "freshness": { "updated": "July 2026", "nextReview": "January 2027" }, "parentArticleSlug": null, "relatedConceptSlugs": [ "what-is-the-comprehension-obligation", "what-is-hybrid-organisation", "what-is-the-abc-framework" ], "relatedArticleSlugs": [ "agents-grading-themselves", "we-fired-an-ai-agent" ], "sources": [ { "id": "S1", "title": "Comprehension Debt: The Hidden Cost of AI-Generated Code", "publisher": "Addy Osmani, O'Reilly Radar", "date": "2026-04-01", "url": "https://www.oreilly.com/radar/comprehension-debt-the-hidden-cost-of-ai-generated-code/", "supports": [ "AI-assisted engineers finished in the same time but scored 17 points lower on comprehension, 50 percent against 67 percent", "output ships while understanding does not", "comprehension debt is scoped to the codebase, not to the operator" ], "method": "Randomised controlled trial, engineers learning an unfamiliar library with and without AI assistance", "n": 52 }, { "id": "S2", "title": "AI Literacy: Questions and Answers", "publisher": "European Commission", "date": "2025-02-02", "url": "https://digital-strategy.ec.europa.eu/en/faqs/ai-literacy-questions-answers", "supports": [ "Article 4 has applied since 2 February 2025", "national market surveillance authorities begin supervising and enforcing on 2 August 2026", "the obligation is to ensure a sufficient level of AI literacy, assessed proportionately, with no method prescribed" ] }, { "id": "S3", "title": "Hybrid Org Doctrine — the Comprehension Obligation clause", "publisher": "MING Labs", "date": "2026-05-01", "supports": [ "the 30-minute unaided reproduction standard, rehearsed quarterly", "a failed rehearsal demotes the agent one autonomy level", "three primary outputs per agent as the scope of the rehearsal" ] }, { "id": "S4", "title": "MING Labs operating record — fleet status and autonomy levels", "publisher": "MING Labs (internal)", "date": "2026-07-01", "supports": [ "seven agents in production under named human owners", "the clause is live in the Chief of Staff agent's job description and rolling across the remaining agents", "no rehearsal-failure rate is published because the rollout is incomplete" ] } ] } ] }