How TeamShared Is Building a Living Map of the AI Frontier
The AI frontier moves faster than a reading list. New companies emerge, products change, and capabilities that looked experimental become practical. Keeping up takes more than collecting links. It takes a process that turns discovery into knowledge the whole organization can use.
That’s why we created AI Startup Research: a recurring research project that brings AI companies, their founders, and their products into TeamShared’s shared ontology.
Our goal is to bring ten AI startups into the ontology every hour, around the clock. Not just ten names in a spreadsheet, but researched companies connected to people, evidence, and context that remain useful after the research session ends.
From occasional research to a standing process
Market research often starts with a question and ends with a document. Someone investigates a category, shares a summary, and moves on. The next person repeats much of the work because the useful context is scattered across browser tabs, chats, and notes.
AI Startup Research makes that work repeatable. Each research cycle identifies an AI-agent company we haven’t covered, checks existing records to avoid duplicates, investigates public sources, and records what can be verified.
The output includes company and founder records, a source-backed dossier, and an entry in the shared company tracker. A daily digest brings the findings together for human review.
The digest is the readable summary. The underlying records are what make the research reusable.
Why an ontology matters
A list tells you which companies exist. An ontology helps describe how companies, people, and products relate to one another.
A startup becomes an organization record. A verified founder becomes a person connected to it. Research adds supported attributes and a dossier explaining what the company does and why it matters.
That structure gives future work somewhere to start. When we investigate agent collaboration, we can return to companies already researched in that area. When a product question overlaps with infrastructure automation, the relevant dossier is available instead of buried in an old conversation.
The distinction is practical: we’re not only saving what an agent read. We’re organizing it so another person or agent can use it.
What the research is finding
On November 6, the project researched eight companies: Revo, Serval, Automat, Kebra, Ando, Empirik, Manufact, and Plato.
Ando’s positioning around messaging for humans and AI agents provides a reference for agent-native collaboration. Empirik’s work on AI-driven infrastructure changes offers a different perspective: how agents move from understanding an engineer’s intent to evaluating and executing a change.
These are different parts of the AI landscape. Recording them through the same research process makes them easier to revisit and compare without losing their individual context.
Over time, the project builds a broader picture than any single day’s digest can show.
Continuous intake needs evidence, not just volume
Ten startups an hour is a throughput goal, not a reason to lower the standard of research.
A company’s marketing claim is not proof of a capability. A founder’s title needs a source. Missing funding information should remain missing rather than become a plausible-looking number.
The process checks for duplicates, keeps sources with the research, and leaves unknowns explicit. Research dates also matter: a dossier describes what was known when it was written, not a promise that every detail remains current forever.
Continuous discovery doesn’t mean complete coverage. It creates an expanding body of evidence that can be revisited, corrected, and extended.
Research that stays with the organization
The most useful outcome isn’t a larger tracker. It’s less repeated work and better-informed decisions.
A strategy discussion can begin with existing research. Product planning can draw on previously investigated companies. An agent can retrieve a dossier rather than reconstruct the same background from scratch.
The process supplies new observations; shared memory preserves the context. Together, they help each research cycle contribute to the next.
This is TeamShared’s mission in practice: agentic business process memory and execution for any organization. In this project, the business process is research, and its output becomes organizational knowledge rather than a disposable chat response.
The aim isn’t to claim we know everything happening in AI. It’s to keep learning systematically—and make sure what we learn remains available to the organization.