Conductor is the execution layer of Taskstreamer. Human and AI work in one record, inside boundaries your organisation set and a human ratified.
Every output has a named owner · Irreversible steps wait for a human · One record, human and AI
Fifty founding-customer slots. A direct call when your name comes up, not a newsletter.
The gap
A developer prompts a model inside their IDE. That model spins up agents of its own. A scheduled job reads tickets on a heartbeat and spawns more to work them. It is fast, it is genuinely productive, and it happens almost entirely inside one person's session.
Nobody else sees it. There is no review, no shared context, and no record of what was asked, what was rejected, or why a particular approach was taken. The output arrives. The reasoning does not. Multiply that across everyone with a licence, at the volume AI makes possible, and an organisation accumulates a large and growing body of work it cannot account for.
That is what actually stops things. Not whether AI works, but who takes responsibility for what it produced. Someone has to be able to say what was made, on what basis, who decided it was acceptable, and whether the organisation stands behind it. Today that answer lives in individual chat histories, if it exists at all.
By the time anyone reviews the output, the agent has already acted. Reviewing it at that point is not oversight. It is incident response.
What it is
Taskstreamer spans three layers. Strategy sets direction and priority. Blueprint turns that into plans, and is built into Conductor rather than sitting beside it. Conductor is where a planned feature is broken into work-items and actually gets done, by people and by AI, in one place.
A work-item is called a Thread, because the work is a conversation a team takes part in rather than a ticket assigned to one person. Because the layers connect, a piece of finished work traces back through the plan to the intent it served. Accountability runs upward as well as downward: not only who produced this and who approved it, but what it was for.
Teams use Conductor the way they use an issue tracker. Threads, owners, phases, dependencies, approvals, and iterations, whether those are sprints or project reporting periods. Agile and project-oriented operating models are both first-class and can be used side by side.
Switch AI on and nothing about the record changes. Same threads, same phases, same approvals, same audit chain. AI works inside the structure the team already uses, and what it produces lands in the same record a person's work would, with the same visibility, the same review, and the same named owner.
What it does
A Thread shows what was asked, what context was used, what the AI produced, what was rejected and why, and who decided. Colleagues can see it, comment on it, redirect it, or take it over. AI stops being a private conversation and becomes work the team can join.
Not a chat history belonging to whoever happened to prompt it. A Thread with an owner, a reviewer, and a record of the decision to accept it. That is the artefact that lets an organisation say it stands behind what it shipped.
Every action is classified as reversible or irreversible, internal or external. AI performs the work up to the boundary of consequence. The irreversible step waits for a named human, or for a policy a named human ratified that permits it to proceed automatically inside a defined envelope.
Conductor does not ask a model to respect a boundary. The credential reachable from the model's execution context does not carry the scope to perform an irreversible action. No prompt, no injected instruction and no compromised model changes that, because the boundary sits in the infrastructure rather than in the wording.
When a decision reaches a human they get a decision-grade briefing: what is about to happen, what it affects, on what basis, where each parameter came from, and what the current value is where something is being changed. A payee account arriving from an inbound email, differing from the account on record, is shown as exactly that. The approval is recorded together with the briefing that preceded it.
Conductor reads your approval hierarchies, signing limits and dual-control settings from the systems that already hold them, and asks a human to confirm each one. Every enforced rule carries its source, the value extracted, the person who ratified it, and when. Constraints tighten locally and never loosen. Widening anything is deliberate, recorded and dual-controlled.
Every step, call, decision and action lands in an append-only chain. A separate component with read-only access checks that chain, and your connected systems, against the guarantees Conductor makes. A system that audits itself audits nothing.
Two ways in
Run Conductor as your work system. It earns its place immediately: structured execution, clear ownership, enforced approvals, dependencies that hold, and a complete record of who decided what.
That first phase does the harder half of the AI work as a side effect. Your organisational model is read from your existing systems and ratified. Action classes are defined. Release policies are written against your real thresholds. Gates and dual controls are exercised by people, on live work, until you know they are right.
Switching AI on is then a switch rather than a migration, and it can be thrown one phase, one team or one action class at a time. That is a very different conversation than approving a model and a control framework at once.
Start with what is actually running. Conductor registers the agents, jobs and configurations in play, what each is permitted to reach, and who owns it.
It then reconciles what your systems of record show against what was authorised, over history, including activity that predates Conductor entirely. That produces a finding: what has been changed and by what, where permissions exceed purpose, and which work has no owner who could answer for it.
From there, work moves inside the boundary team by team, highest consequence first. The shadow usage does not need to be banned. It needs somewhere better to go.
Anything claiming to wrap an existing agent estate without moving execution is selling observation and calling it control.
Why it is different
| Category | What it governs | What it produces |
|---|---|---|
| Issue trackers | Human work | A record with a hole where the AI work happened |
| GRC and policy tooling | Documentation of AI systems | A register, and false confidence |
| Model governance platforms | Models you have deployed | Inventory, monitoring, explainability |
| Agent platform native controls | The vendor's own agents | Logs, and lock-in |
| Conductor | The work itself, human and AI | Accountable output, and the evidence behind it |
Everything else observes work happening somewhere else and reconstructs it from logs, which is why coverage always has holes where tools hand off to each other. In Conductor the work and its governance are the same record, so the chain is complete because of how it was produced. It is also why one rule covers a person and an agent alike, and why nobody has to maintain a separate account of what the AI did.
Your thresholds are already in your ERP. Your dual controls are already in your payment platform. Conductor imports them, proposes a model, and asks a human to ratify it. Onboarding is a confirmation exercise, not a policy authoring project.
What it is worth
What we do not claim
Conductor documents in writing what it does not detect and cannot guarantee. You can read it before you buy, not after.
Conductor is in build and opening to founding customers. The Strategy and Planning layers of Taskstreamer are in production today. The full list of what we do not detect and cannot guarantee is on the limits page.
Who this is for
Any organisation where AI is already producing real work and nobody can say who is accountable for it. That is most organisations with engineering, operations or back-office teams, and it arrives sooner than expected: the first agent with write access to a real system is the moment the question becomes concrete.
The need bites hardest where consequences are formalised, in financial services, insurance, healthcare, pharmaceuticals, energy and regulated manufacturing, and wherever work touches a system of record.
We will show you what is running today, what it would take to make it accountable, and what we could not govern, before you commit to anything.
Fifty founding-customer slots. A direct call when your name comes up, not a newsletter.