Why we started Brackett

A year ago, when I was working out what I wanted to build, I went back to my developer roots. A developer opens a brace (or, a bracket) , writes some logic and then closes the brace. That is the entire contract of software, and it is the reason enterprises ever trusted it. What you wrote is what ran. The code did not wake up one morning and rewrite itself. It did not go looking for work nobody asked it to do. It stayed inside the scope it was given, every time.

That is what the Brackett name implies. It is the coding paradigm brought to AI. Meaning Bracket your Tasks and gives the power to every business user to do so. When you teach a machine a complex task, the scope of that task should be as explicit as a pair of braces, and the behavior inside it should be as predictable as software. Agents serve humans. The human is the one who gets elevated. Not the other way around.

That is why we started Brackett and we stay true to these principles and offer it to our customers through our platform.

The thesis

Look at how operations actually run in the physical economy, where we focus. Manufacturing, logistics, fleet, freight and energy. The work that consumes the day is repetitive, manual and reactive by construction. A warranty claim that needs analyzing. Predictive maintenance on a fleet. A change order that reaches a line which has already run. A supplier risk that nobody accounted for. A shipment sitting in customs that could have been avoided.

Two things make it harder here than anywhere else. The first is variability. Real world conditions do not hold still, and no two exceptions arrive in the same shape. The second is that the systems, processes and tooling meant to help the operator are disconnected from each other, so the work of holding them together falls on people.

Then there is the part nobody has solved. The procedures, the judgment and the exceptions by which a business runs are not in any system. They are scattered across tools and inboxes, and in the tacit knowledge of people doing the work. I call this the unwritten dataset. Every operations team on the planet has one. None of them own it.

Here is what it costs. A freight shipment gets held in customs after a sudden reroute. A team scrambles across disconnected tools, chases people, runs handoffs, and eventually arrives at a decision: add it to the clearance queue, which a broker will handle in bulk next month. The ERP records the outcome. How anyone got there is gone. The next time it happens, the work starts from zero.

Previous generations of automation addressed the manual work but failed to turn that into an intelligent asset that the company can build and rely upon, partly due to the limitations of the technology layer underneath. Now that has changed and that’s what gets me super excited about the intelligent foundation and the possibilities it can offer to customers, when done right.

Why the current architecture needs a rethink

The breakthroughs of the last two years give us a real opening to rethink how this work is done. The architecture most agents are built on will need to be thought through as well. The main reasons are:

Problems get solved in silos, so nothing one agent learns is available to the next. Judgment is rarely captured, and when it is captured, it is not retained in a way that survives the next iteration, the next vendor or the next model version. The way forward is humans and agents working together and learning from each other continuously, and almost nothing is built for that. And a business has no way to own the recipes of success that make its playbook better than its competitors.

Brackett is built for those four gaps. Agents tailored to the business, handling parts of the work without loss of fidelity, automating the decision as well as the task, and doing second order thinking on both the output and the process itself. That is only possible because of the architecture underneath: a connected agentic workforce reasoning over one shared account of how this business works.

Our vision is being built in 3 phases.

We have built the technology that lets an operator teach a complex workflow directly, or lets Brackett infer the patterns that repeat in day to day operations. The unwritten dataset gets materialized alongside the semantic understanding of the domain and its ontology, from the work itself, with nobody stopping to document anything.

Those patterns become a connected agentic organization that can be trusted, measured and improved as the business changes.

Finally, Every organization gets a local small model tuned on its own operating knowledge, so investigations run faster, long analyses become possible, and processes get redesigned on evidence rather than on memory.

Agents you cannot measure are agents you cannot trust. So, we built the Agent Effectiveness Index and open sourced it. This index needs a community of people and organizations to come together and you can find the reference to it here.

Where this goes

IDC expects more than a billion agents in the enterprise by 2029. Brackett is bringing agents too. Ours differ in one specific way: they sit on a foundation built for connected learning, retained memory and deterministic execution, which is what lets them form a mesh that finally records the unwritten dataset and compounds on it.

Models scale on data. Brackett scales on procedures. We are live today. Come find us at brackett.ai.

Cheers.