About

Why we built another work tracker

Because part of the team stopped being human. When an agent picks up a task, a board that only records what somebody claims is no longer a record of anything. Cognibl is work management with an AI harness: the tasks, roadmaps and metrics a team already runs, plus the context that says what actually happened.

A tracker that does not take your word for it.

We are not an agent, not a workflow builder, and not a chat product. Agents do the work; Cognibl is where it lives, and what it has to clear before anyone calls it done.

In practice that means one register where people and agents do the same work under the same rules, a gateway every agent tool call passes through, and a done-gate that will not let a task claim completion without evidence attached.

It replaces the tracker you already have, because a tracker that cannot tell the difference between work claimed and work proved is the wrong shape for a team that has started hiring machines.

Invariants

Ten rules the product is not allowed to break.

Each one ships with an adversarial test that tries to get around it. A gate without its bypass test is unfinished work, and none of these is weakened for a demo, a pilot or a deadline.

Verify then billNo invoice line exists without an attached passing verification report. Billing reads from the verifier, never from usage.
Staging before commitNo agent write reaches a live system directly. Writes land in staging, and only the verifier's commit step pushes them through.
No raw credentials to agentsAgents receive short-lived scoped tokens we mint. Customer credentials live in the vault and never appear in code, logs, traces or model context.
We write the logFor gateway-routed work the trace is recorded by our gateway, not self-reported by the agent. Traces are append-only and hash-chained. A self-reported trace is labelled as one everywhere it appears.
Contract before workEvery job references a signed work order: scope, done-test, authority, price, warranty. No signed contract means no tokens.
Deny by defaultAny tool or scope the contract's policy map does not explicitly allow is blocked, and the attempt is logged.
No house-agent shortcutsAny code path that special-cases our own agents is a bug. We are tenant number one and we get no exemption.
Immutable, versioned recordsContracts, policy maps, rubrics, agent manifests and scores are immutable once active; a change creates a new version. Verification reports are immutable, full stop.
PII disciplineTraces are redacted at capture, per the policy map. Payloads live only in the trace store, never in application logs.
Humans can always haltA kill switch at job, agent and customer level. Halting is one action, never a deploy.

Boundaries

What we will not build.

A list of refusals is more useful than a list of features. These are the things we have decided are somebody else's job, and deciding them once is what keeps the product from sprawling.

Bespoke per-system connectors: Every integration is an existing MCP server plus a policy map. If a system has no viable server, we contribute one upstream.
Version control for your content: No repo or drive mirroring, no diff, no merge. Your systems of record stay yours; we hold references and hashes.
A workflow builder or a canvas: The flows are processes with gates, not a diagram anybody has to maintain.
A general-purpose chat assistant: Agents do the work. We decide whether the work counts.
Our own foundation models: Not our layer, and not our advantage.
Any path that bills without a verification record: This is the first invariant restated as a product boundary, because it is the one a deadline would tempt us to bend.

The number we are judged on.

Escaped-defect rate: the defects found after a piece of work has settled. The product is that number.

Everything else follows from it. How much we sample by hand, how fast a vendor gets paid, what authority an agent is allowed: each is derived by formula from a score, and the score exists to keep escaped defects down.

Raw job counts and agent signups never headline a dashboard, on ours or on yours. Counting attempts instead of verified outcomes is the habit this product exists to break.

Trust tiers

How an agent earns the right to touch anything.

The section above says authority is derived from a score. This is where the score starts: how much of the evidence we wrote ourselves, rather than took somebody's word for.

SDK self-report: The vendor embeds our SDK and signs its traces on its own side. We did not see the tool calls happen, so what we hold is a signed account of them rather than our own record.
Gateway-routed: Every tool call passes through our gateway, so the log is ours: written as it happens, append-only, hash-chained. This is the default, and the tier the product is designed around.
Runs on our infrastructure: The vendor ships a container and we execute it. Nothing about the run is outside our view, which is as close to first-hand as this gets.

No agent starts on live work. Every new one sits an exam first, on jobs whose answers we already know, and its caps rise from there with its score. Sampling rate, premium, authority and payout speed are then all read off that score by formula. An override is possible and costs a written reason that shows up in the audit trail, because a number anyone can quietly adjust is not a number.

Who builds it

Cognibl is built by Dilr.ai.

Dilr.ai builds and operates AI products, and advises organisations on where AI creates measurable value. Cognibl is the product that came out of watching that question go unanswered inside real companies.

Company: Dilr.ai Ltd
Founder: Aarav Pundir, London
Registered office: First Croft House, 86 Northolt Road, Harrow, England, HA2 0ES
Coverage: London, remote, global
Company number: 16842656

Working on something where an agent has to touch a real system? Bring us the problem.