Decawork
Lets IT and security teams see, approve and monitor every internal AI agent employees build, instead of agents spreading unchecked across the company.
🔗 Visit Decawork
Description
Once a company's employees start building their own AI agents in tools like Claude Code or Codex, IT quickly loses track of who has access to what, which agent can touch sensitive data, and what happens when one misbehaves — a problem people are starting to call "agent sprawl." Decawork acts like a control tower for that: agents get deployed through it rather than in the shadows, so IT can approve, watch and shut down agents the same way they already manage employee laptops or software access.
Under the hood, Decawork is a governance/control plane where agents built in Claude Code or Codex get deployed with scoped credentials, go through an IT approval workflow before going live, and generate a full audit log of every action they take, with automated alerting on failures or suspicious behavior. It's a Y Combinator S26 company founded by people with backgrounds in AI compliance (Barclays) and AI systems engineering (NVIDIA), which fits the specific niche it's targeting: the research cited on its own launch claims only 1 in 7 companies currently deploy AI agents with any IT approval process at all — Decawork exists to close that gap.
💬 Our review
The short version: Decawork is a bet on a problem that's about to become much more common — engineering and ops teams have started spinning up AI agents faster than IT can track them, and Decawork is one of the first dedicated tools built specifically to govern that, rather than repurposing generic IAM software.
The strength is specificity: scoped credentials per agent, mandatory approval workflows, and full audit logging are exactly what a security team needs to answer "which agent touched what, and who approved it" — a question that's currently unanswerable at most companies deploying agents informally.
The honest caveat: this is a brand-new YC S26 company with no long public track record, undisclosed pricing, and (at fact-check time) an inconsistently reachable homepage — treat it as an early, promising bet rather than a proven enterprise tool, and pilot it on a small team before rolling out company-wide governance on top of it.
💰 Pricing
📊 Global score
🤖 AI-enriched data
Tarification non publiée au lancement (août 2026), probablement B2B sur devis.
Pros
Cible une lacune réelle et récente : la prolifération d'agents IA non supervisés (agent sprawl)
Identifiants scoppés par agent + workflow d'approbation IT avant mise en production
Journalisation complète des actions de chaque agent
Fondateurs avec expérience en conformité IA (Barclays) et ingénierie IA (NVIDIA)
Cons
Entreprise très jeune (YC S26), pas d'historique long terme
Tarification non publiée
Site principal instable/bloqué par moments au moment de la vérification