Blog

Guides on AI agents and proof of work

Practical guides for teams where people and AI agents share the work: how to say what done means, and how to check that it happened.

RSS

21 September 2026Shadow AI

How to govern shadow AI agents you cannot see

A shadow AI agent is one running in your company that nobody approved. Enterprises are heading for over 1,600 each, and most cannot govern the ones they have.

4 min

21 September 2026ROI

How to prove ROI on AI agents when the board asks

Most teams know what they spend on AI agents but not what they get back. Boards now ask for auditable outcomes, not user counts. This is how to show them.

3 min

21 September 2026MCP security

What MCP tool poisoning is and how to stop it

MCP tool poisoning hides instructions in a tool description that the agent reads and the user never sees. Benchmarks show it working on most agents tested.

3 min

3 September 2026AI pilots

Why most AI agent pilots fail and what works instead

MIT found 95% of enterprise GenAI pilots delivered no measurable profit. The model is rarely the problem. This is what the successful 5% do differently.

3 min

3 September 2026Agent washing

How to tell a real AI agent from agent washing

Agent washing means calling a chatbot or a script an AI agent. Gartner expects it to help cancel over 40% of agentic projects by 2027. This is how to spot it.

3 min

3 September 2026Agent identity

How to secure AI agent identities when they outnumber your people

Machine identities now far outnumber people, and most IAM tools cannot manage AI agents. Treat each agent like a contractor with limited, temporary access.

3 min

21 August 2026Control

Where to put a human checkpoint on an agent

Full autonomy is the wrong goal. Let an agent run through reversible steps and stop it before anything that cannot be undone: sending, deleting, charging.

4 min

21 August 2026Metrics

What replaces velocity when agents write code

Story points measure effort, and an agent can produce a hundred in an hour. Better measures are verified throughput, first-pass rate and where the time goes.

4 min

21 August 2026Standards

What is AGENTS.md and what goes in it?

AGENTS.md is a README for agents: the build commands, conventions and boundaries an agent needs. Over 60,000 repositories ship one. What to put in yours.

4 min

21 August 2026Review

How much agent work should a human review?

Teams stop reviewing everything long before anyone decides to. It is better to sample on purpose, at a rate set by evidence, than to sample by accident.

5 min

21 August 2026Compliance

EU AI Act logging: what agents must record

Article 12 has applied to high-risk systems since 2 August 2026. It requires automatic, lifetime event logs, which rules out a self-reported trace.

4 min

19 August 2026Context

Why AI agent failures are context failures

Agents rarely fail because the model was not capable. They fail because nobody told them what done meant, and the 2026 data now shows this clearly.

4 min

19 August 2026Proof of work

What is proof of work for AI agents?

Proof of work here has nothing to do with mining. It is the evidence a task carries to show the work was done, and the reason a status can be refused.

3 min

19 August 2026MCP

What is an MCP gateway and when do you need one?

An MCP gateway sits between your agents and your MCP servers. It issues scoped tokens, enforces policy and writes the trace. This is when you need one.

4 min

19 August 2026Metrics

Measuring agentic AI: what to instrument

Most teams running agents in production have no way to measure them. This is the short list of what to track, and of what you can safely ignore.

4 min

19 August 2026Security

How to give an AI agent GitHub access safely

Do not hand an agent your personal access token. Compare deploy keys, fine-grained PATs and GitHub Apps, then scope, stage and log every write it makes.

5 min

19 August 2026Metrics

What first-pass verification rate measures

First-pass verification rate is the share of agent tasks that pass on the first try. It measures how clear your spec is more than how good your agent is.

4 min

19 August 2026Metrics

Cycle time for agent work: the four phases

One cycle-time number hides where the time went. Split it into spec, build, verify and settle, and the bottleneck in agent work becomes clear.

4 min

19 August 2026Comparison

Cognibl vs Linear for teams that use AI agents

Linear is faster and better designed than we are, and its workflow is deliberately fixed. The difference is what happens when an agent says a task is done.

4 min

19 August 2026Comparison

Cognibl vs Jira: proof before a status change

Jira can block a transition with validators. The difference is what gets checked: field input you configure, or evidence against a definition of done.

4 min

19 August 2026Category

Agent governance vs observability vs verification

Three tools that sound alike but answer different questions: what happened, what is allowed and whether the work is acceptable. Only one of them can block done.

4 min

18 August 2026Definition of done

Writing a definition of done an AI agent can pass

A definition of done written for a person is a reminder. Written for an agent it is a specification, and the difference shows up in your first-pass rate.

4 min

17 August 2026Verification

How to check what an AI agent did

Reading every diff does not scale, and spot checks miss what matters. Use three layers instead: machine checks, a model review and a human sample.

4 min

Guides on AI agents and proof of work

How to govern shadow AI agents you cannot see

How to prove ROI on AI agents when the board asks

What MCP tool poisoning is and how to stop it

Why most AI agent pilots fail and what works instead

How to tell a real AI agent from agent washing

How to secure AI agent identities when they outnumber your people

Where to put a human checkpoint on an agent

What replaces velocity when agents write code

What is AGENTS.md and what goes in it?

How much agent work should a human review?

EU AI Act logging: what agents must record

Why AI agent failures are context failures

What is proof of work for AI agents?

What is an MCP gateway and when do you need one?

Measuring agentic AI: what to instrument

How to give an AI agent GitHub access safely

What first-pass verification rate measures

Cycle time for agent work: the four phases

Cognibl vs Linear for teams that use AI agents

Cognibl vs Jira: proof before a status change

Agent governance vs observability vs verification

Writing a definition of done an AI agent can pass

How to check what an AI agent did

Keep reading