The StackAI Manifesto

Mar 27, 2026

Who We Are

We're Antoni Rosinol and Bernard Aceituno. We met at MIT, where we both completed our PhDs in AI research. Between us, we'd spent more than seven years pushing the boundaries at Facebook’s AI Research team and NASA's Jet Propulsion Laboratory.

We were at MIT when the world changed.

Foundation models arrived, and we saw immediately what they would become: transformational for the enterprise. For the first time, machines could handle uncertainty. They could interpret unstructured data (messy documents, ambiguous emails, complex forms) with a level of judgment that was previously impossible to automate. Processes that had been stuck in manual workflows for decades were suddenly within reach.

But we also saw something that others were ignoring.

These models were stochastic. Non-deterministic. Brilliant at judgment but dangerous when left ungoverned. And the industry's answer was to throw language models at everything: give them a task, hope for the best, and call it automation.

We knew that wasn't how the world works. Not in banking, defense, healthcare, or any environment where a wrong decision has real-life consequences.

So we moved to California, joined Y Combinator, and founded StackAI to build something different.

What We Believe

We believe that language models have unlocked something extraordinary: the ability to automate processes that couldn't be automated before. Now, AI can realistically handle tasks that involve judgment, interpretation, and reasoning over unstructured data. That's a stunning breakthrough.

But here's what most AI companies get wrong:

You cannot just throw a language model at a business-critical process and expect it to run.

Giving AI complete autonomy should not be the goal. The safe, responsible method is to break down safety-critical and business-critical processes into well-defined, deterministic steps and then give language models the specific tasks that require their unique capabilities: judgment, analysis, and interpretation of unstructured data. Each model operates within a structured workflow, with a defined set of tools, clear boundaries, and human oversight at important junctions.

Determinism is the container. Autonomy lives inside it.

This is how the enterprise of the future will operate, not through black-box magic, but through disciplined processes where AI handles what AI is best at, humans handle what humans must, and every step in between is traceable, auditable, and governed.

What This Looks Like in Practice

One of the largest banks in the world uses StackAI to handle anti-money laundering investigations. Here's how it works:

An alert arrives. The central bank sends an email flagging a potential money laundering case.
An AI agent parses the notice. A language model reads the unstructured document, interprets the details, and extracts the relevant information: names, accounts, and indicators.
The agent investigates. Autonomously, it queries the internal customer database, runs web searches, aggregates data from multiple sources, and researches whether the flagged individual is a customer and whether the evidence supports the claim.
A judgment is made. The language model assesses the findings and determines whether the case warrants account closure.
A human decides. Before any action is taken, the workflow surfaces the recommendation to a compliance officer through a human-in-the-loop step. A real person reviews the evidence and approves or declines the request.
The agent acts. If approved, the language model simulates a browser session, enters the bank’s internal portal, and closes the account.

Every step is predefined; the process is deterministic. But within that structure, the language model exercises real autonomy by interpreting documents, conducting research, making assessments, and filling forms. Then, the human makes the final call.

This is what we call an Agentic Workflow. Not AI running wild, AI is running right.

What We're Building

StackAI is the enterprise platform for building, deploying, and governing custom AI agents with Pentagon-level security, in minutes.

We give organizations the power to design agentic workflows: systems where AI agents operate together within structured processes, each handling a specific task.

At the heart of StackAI is a no-code, drag-and-drop building platform. This canvas democratizes AI for the entire enterprise: business users can create and orchestrate agentic workflows just as easily as your best engineers.

This means teams can parse documents, query data, and flag compliance gaps without lifting a finger, but they are brought into the loop before sensitive actions are taken. When the request for approval is completed, agents execute precisely and reliably at scale.

We deploy anywhere: your cloud, on-premises, air-gapped, or in our multi-tenant environment. We send our own forward-deployed engineers to work alongside your teams, designing agents that map to your actual processes, systems, and security requirements.

And we wrap it all in a governance layer that gives IT teams and CIOs full visibility: what agents are running, where they're hosted, what models they use, how they're built, what tools they access, who has permissions, and how every one of them can be edited, managed, or locked.

The Enterprise of the Future

Every process is owned by an agent; every agent is governed by the enterprise.

We envision a future where every manual, tedious component of the enterprise is powered by an AI agent, and every agent is orchestrated by the team that owns it.

Not a future where AI replaces humans, but serves them so they can do their best work. Where compliance teams have co-pilots they actually trust. Where government agencies no longer wait for every task to be run manually. Where regulated industries finally move at the speed of technology without compromising on safety.

A future where no one has to "just trust the AI" because the AI is built to earn that trust, step by step, decision by decision.

Our Partners

We are proud to have earned the trust of enterprises like BAE Systems, Nubank, and LifeMD to deploy AI agents at scale across their operations.

They chose us because in their highly-regulated industries, there is no room for error, no tolerance for opacity, and no shortcut to safety.