FAQ

Why do enterprise AI agents need governance and human approval points?

Because an agent differs from ordinary software in one decisive way: it chooses its own next actions, and its errors amplify at machine speed. Governance answers four questions — which data it may touch (permissions), which actions require a human (approval points), what it did (audit trail), and how it stops (kill switch). Approval points are not distrust of the technology; they keep accountability on the human side: external sends, money, deletions and commitments pass a human, everything else flows automatically. The EU AI Act likewise requires human oversight for higher-risk systems — governing early is the cheapest compliance path.

Figure 01 · Three approval grades, by irreversibility

  1. 01

    Grade 1: Fully automatic

    Reversible, low-impact, internal only

    Examples: retrieval, drafts, internal summaries Oversight: logs + sampled review
  2. 02

    Grade 2: Human spot-check

    Externally visible but retractable, medium impact

    Examples: sending drafted replies, bulk internal notices, bulk CRM updates Oversight: sampled checks; a failed check pauses that class of action
  3. 03

    Grade 3: Approval before action

    Irreversible, externally binding, or touching money or production data

    Examples: sending email, changing production data, money movement, quotes and commitments Oversight: per-action approval; two-person approval for the highest-risk actions
Any action not on the table defaults to Grade 3. The table is final before deployment; new tools, data sources and external channels are graded before switch-on.

The speed of errors changed

A traditional software bug is static: one defect, one feature affected. An agent's error is dynamic: one wrong judgment triggers a chain of follow-on actions, and by the time a human notices, ten steps have executed. Governance does not aim for a mistake-free agent; it aims for mistakes that cannot amplify and can be rolled back.

A typical scenario (illustrative, not a real client case): a support agent misreads "considering a return" as "confirmed return", raises a refund, emails the customer confirmation, updates inventory, and syncs the record to finance. Each step looks compliant alone; together they are a refund that should never have happened. An agent without approval points does not make one mistake; it makes a whole chain of them.

Where approval points belong: three grades by irreversibility

The yardstick is one question: if this action is wrong, can it be undone at no cost? Three grades follow (the grading table is the figure on this page):

  • Grade 1: fully automatic. Reversible, low-impact, internal-only actions — retrieval, drafts, internal summaries. The agent runs them freely; humans sample the logs afterwards.
  • Grade 2: human spot-check. Externally visible but retractable, medium-impact actions — sending drafted replies, bulk internal notices, bulk CRM updates. The agent acts; humans spot-check a proportion, and a failed check pauses that class of action.
  • Grade 3: approval before action. Irreversible or externally binding actions — sending email, changing production data, moving money, quoting or committing to customers. The agent prepares; a human approves before anything executes.

Two supporting rules: the table is fixed before deployment and anything not on it defaults to Grade 3; and grading never ends — new tools, data sources and channels are graded before switch-on.

Approval is a release, not a brake

The common misunderstanding is that governance slows agents down. Without grading, a company has two options — let everything through (an incident waiting to happen) or have humans watch everything (which throws away the agent's speed advantage). With grading, low-risk actions run faster — nobody approves each small thing — and human attention concentrates where it is genuinely needed.

Audit is not for blame

A complete log — inputs, tool calls, outputs, approvals — exists for review and improvement: which task types fail most, which rules block legitimate work. An agent without logs cannot be iterated, only guessed at.

At minimum, log five things: the input, tools called and their parameters, the output, the approval grade and approver, and timestamps. Fix retention at pilot design time — too short and incidents are uninvestigable, too long and costs grow. Logs also carry compliance value: the AI Act requires high-risk systems to record events automatically (Art. 12), and GDPR requires records of processing (Art. 30).

The compliance dimension: three laws, three jobs

GDPR governs the data. When an agent processes personal data, the controller needs a lawful basis (Art. 6), data protection by design (Art. 25), processing records (Art. 30), appropriate security (Art. 32) and advance assessment of high-risk processing (Art. 35). Note Art. 22: solely automated decisions about individuals, including profiling, are restricted. Fines for breaching the core provisions run up to €20 million or 4% of worldwide annual turnover, whichever is higher (Art. 83(5)).

The AI Act grades systems by risk. High-risk systems require a risk management system (Art. 9), technical documentation (Art. 11), automatic logs (Art. 12), transparency information for deployers (Art. 13), human oversight (Art. 14) and post-market monitoring (Art. 72); deployers must assign oversight to natural persons with the competence, training and authority to do it (Art. 26). Conversational support is usually not high-risk, but users must know they are talking to AI (Art. 50). Fines run up to €35 million or 7% for prohibited practices and €15 million or 3% for other infringements (Art. 99).

NIS2 governs cybersecurity in critical sectors. It applies to essential and important entities in the Annex I and II sectors that meet the size threshold: management bodies must approve the cybersecurity risk-management measures and oversee their implementation (Art. 20), and the measures include supply-chain security (Art. 21(2)(d)) — covering the agent's model and tool suppliers too. Check scope against the annexes and the size threshold; when in doubt, the directive text and national law decide.

All three point at the same evidence: proof that you are in control. The governance framework is that proof, for regulators and customers alike.

What "enough governance" looks like

A workable minimum: the grading table is final and covers every action, approvers are named with cover, the five log fields are captured with fixed retention, and a monthly review covers error rates, overrides and legitimate work blocked.

Bring the agent solution into your workflows

Read the answer — the next step is judging it against your scenario. Discuss an agent solution: tell us the business problem and your systems for a preliminary suggestion.

Related questions

How is GEO different from SEO?

SEO optimises a page's rank in a list of links, aiming for clicks. GEO (Generative Engine Optimization) optimises how discoverable, understandable, trustworthy and citable a company is inside AI-generated answers — the goal is to be in the answer itself. They share technical foundations (crawlability, structured data, content quality), but GEO additionally requires governed entity facts, a machine-readable surface (llms.txt, Markdown mirrors) and a third-party mention ecosystem. A site that ranks well is not automatically visible to AI: engines cite the companies whose facts they can fetch, parse and verify. The concrete differences sit in four dimensions — goal, object, content structure and evaluation — compared item by item in the body below.

How can Chinese companies going global build AI search visibility?

For Chinese companies going global, the visibility gap is rarely content volume — it is the machine-readable surface. Our audit of 30 Chinese outbound companies found machine-readability the weakest category (69.2/100 average): 17 had no JSON-LD on the homepage, 15 had a missing or thin sitemap, 13 had a missing or duplicated H1. The path has four steps: audit bot reachability and machine-readability first; govern entity facts so Chinese and English statements agree; publish question-shaped content built on real overseas buyer questions; then earn third-party mentions (media, communities, partners) and verify with periodic resampling against a fixed question set.

What should a corporate website change for the AI search era?

Five changes. (1) Bot reachability: robots.txt explicitly allows AI crawlers and live fetches are not stopped by firewalls or challenge pages. (2) A machine-readable surface: an llms.txt fact file (optional aid), a complete sitemap, Markdown versions of core pages (optional). (3) Entity structure: JSON-LD naming entity types such as Organization, clean title/H1 discipline, hreflang pairing on bilingual sites. (4) Question-shaped content: pages organised around buyer questions, short answers first, FAQ schema attached. (5) Measurability: analytics that separate human and bot traffic and attribute AI-referred inquiries. Negency's own site was rebuilt to exactly this standard, and the public audit tool scores against it.

What is llms.txt, and should every company website have one?

llms.txt is a plain-text brief at the site root written for AI systems: who the company is, what it does, where the core pages live, and the canonical facts and definitions. It does not by itself raise citation rates — in Ahrefs' May 2026 study of 137,210 domains, 97% of published llms.txt files received zero requests that month, so its direct effect is near zero — but it is the cheapest fact-governance move available: one file anchors a single version of the truth that crawlers can rely on. Verdict: worth adding, useless alone; it works as part of entity-fact governance, structured data and question-shaped content. Negency's llms.txt is a full fact bible: company definition, offering boundaries, sourced cases and citation rules.

What should an enterprise prepare before deploying AI agents?

Four preparations, and we advise against starting without any of them. (1) Scenario selection: pilot on a workflow that is frequent, comparatively rule-clear and verifiable — not the most glamorous one. (2) Knowledge and data foundations: govern the knowledge base first (deduplication, versioning, permissions); a dirty knowledge base only scales errors. (3) Governance design: permission boundaries, human approval points and audit logs decided before launch, not after. (4) A measurement baseline: record the manual process's time and error rates before the pilot, or value can never be proven afterwards. Negency's method is a small controlled pilot, verified before scaling, with governance as a design input rather than a patch.