Midfleet

Govern autonomous agent work from intent to evidence

9 followers

Govern autonomous agent work from intent to evidence

9 followers

Visit website

AI Infrastructure Tools

Giving your agents the infrastructure to communicate, act and regulate real work. From intent to completion with evidence. Midfleet keeps intent, scope, rules, ownership, approvals, blockers and evidence attached to autonomous agent work while agents decide how to get it done.

Free

Launch tags:Productivity•Developer Tools•Artificial Intelligence

Launch Team

Maker

📌

Hey Product Hunt — I’m Ahmed, founder of Midfleet. Autonomous agents can move work forward, but the team remains accountable for the result. That creates a new operating problem. When work crosses agents and people, someone still needs to be able to answer: - Who owns the outcome? - What is inside and outside the approved scope? - Where is a decision or approval waiting? - What is blocked, and who needs to act? - Who owns the next handoff? - What evidence supports completion? We built Midfleet to make those answers visible from intent to completion. Midfleet’s center of gravity is operational control. It goes beyond routing tasks, inspecting traces, or connecting agents to tools. It keeps ownership, scope, approvals, blockers, handoffs, and evidence visible around the outcome as work moves. Today, Midfleet is available self-serve. You can start on the free plan without a credit card: https://midfleet.ai/register?utm... A concrete first test: take a GitHub epic or issue already queued in the Midfleet run queue from accepted to merge—while keeping the human owner, scope, approvals, blockers, handoffs, and completion evidence visible along the way. If you work with agents today, where does control become weakest: ownership, scope, approvals, blockers, handoffs, or evidence? I’ll be here throughout the launch to answer questions and learn from what you try.

Report

2mo ago

BetterClaw

@smashadani Congratulations on your launch, here to support you!!

Report

2mo ago

Maker

@better_shaya Thanks Shaya!

Report

2mo ago

Maker

I have read through a lot of threads lately on PH and I see that we keep using “agent governance” to mean permissions, approval gates and audit logs.

Sure these matter, but they govern individual actions. They do not govern the work.

A workflow can prescribe every step when the path is known. Complex work is different and thats something I discovered when Raju and I were building Midfleet.

The agent needs to investigate, make trade-offs, recover from dead ends and sometimes change its approach.

If we sequence every decision, we have built automation with an LLM inside it. If we simply tell the agent to achieve a goal, we have delegated without boundaries.

The middle ground is a form of mandate:

the intent behind the work
what is in and out of scope
rules that must remain true
who owns the current state
when the agent must escalate
the outcome and evidence required to finish

The agent remains needs to be free to reason inside that space. Governance does not tell it how to think. It makes the authority under which it acts explicit and durable across agents, humans and handoffs.

A workflow says what happens next.

Governance says what the agent is allowed to decide.

How are you encoding that boundary today without turning autonomous work back into a rigid sequence? Keen to hear thoughts on this :)

Report

2mo ago

Forum Threads

p/midfleet

•

1mo ago

An agent said “done.” Midfleet said “not yet.”

The pull request reached review, but three required evidence items were still missing and two checks needed attention.

So the handoff stayed blocked.

p/midfleet

•

1mo ago

The agent should not be the durable unit of work. The work should be.

We keep focusing on making agents smarter and more autonomous. Frankly the more I have worked on building midfleet the less I worry about what model the agent is using.
I think the harder problem was making the work survive the agent. I.e. I needed the mandate I gave the agent, to survive regardless of what happens to it.
Agents will stop. They ll hit limits, crash, lose context, or get replaced by a better model.
Completely normal.
The goal, owner, boundaries, decisions, progress and proof of completion should still be there.
Then one agent can implement.
Anpther one can review.
A human can approve.
Another agent can finish.
The worker can change without the job becoming an orphaned chat.
Agents should be replaceable. Responsibility shouldn t be.

p/midfleet

•

2mo ago

What breaks first when autonomous agent work changes hands?

Midfleet came out of our own frustration with multi-agent work. We were spending too much time babysitting runs, repeating context and pulling agents back toward the original intent.

Tools like Temporal, Galileo and Factory solve useful parts of the stack. The gap for us was keeping the work itself governed as it moved between agents and people.

View all

I have read through a lot of threads lately on PH and I see that we keep using “agent governance” to mean permissions, approval gates and audit logs.

Sure these matter, but they govern individual actions. They do not govern the work.

A workflow can prescribe every step when the path is known. Complex work is different and thats something I discovered when Raju and I were building Midfleet.

The agent needs to investigate, make trade-offs, recover from dead ends and sometimes change its approach.

If we sequence every decision, we have built automation with an LLM inside it. If we simply tell the agent to achieve a goal, we have delegated without boundaries.

The middle ground is a form of mandate:

the intent behind the work
what is in and out of scope
rules that must remain true
who owns the current state
when the agent must escalate
the outcome and evidence required to finish

A workflow says what happens next.

Governance says what the agent is allowed to decide.

How are you encoding that boundary today without turning autonomous work back into a rigid sequence? Keen to hear thoughts on this :)