Every enterprise AI story in 2026 has been about deploying agents. The story nobody is telling loudly enough is what happens when you have dozens of them.

Thoughtworks addressed that gap directly at the Databricks Data + AI Summit in San Francisco this week, launching Agent/works — a platform that gives enterprise teams a single control plane and a governed runtime for every AI agent they run, regardless of which cloud it lives on.

The announcement dropped on June 16 as 30,000+ data and AI professionals filled the Moscone Center. The timing is deliberate. Governance is the theme of the summit, not capability. Everyone knows agents can do things. The question is whether organisations can track what they are doing, what it is costing, and whether it is safe.

What Agent/works Actually Does

Agent/works operates as a foundational layer. Enterprise product teams build their custom agentic applications on top of it — they do not replace it. The platform provides three core things:

A governed runtime. Every agent runs inside an environment where policies are applied and can adapt during execution. This is not a static ruleset applied at deployment. The governance is live, adjusting as the agent operates.

A multi-model backend. Teams register any AI model using a standard API, connect any tool, and route work to the right model for each task. The platform connects to cloud-native services and third-party agents through standard interfaces, meaning you are not locked into a single provider or architecture.

A centralized registry. This is where finance teams and CIOs will pay attention. Every agent, model, tool, and policy appears in one place, with evaluations, usage analytics, and cost controls visible in a single view. You can see what every agent is doing, how much it is costing, and whether it is behaving within defined boundaries.

Thoughtworks has been running its own agentic development platform — AI/works — on top of Agent/works in production. That means this is not vapourware. The company is dogfooding the governance layer they are selling.

The Problem This Solves

The numbers explain why this matters. Enterprises are now running an average of 12 AI agents, according to the Belitsoft 2026 AI Agent Report, with 57 percent of companies having agents in production. But many of those agents still operate in isolation from each other and from any centralised oversight.

The result is a governance gap that grows with every agent you add. You might know what one agent is doing. You probably cannot answer, across your entire organisation, who authorised which agents to access which data, how much was spent this month on AI inference, or what the complete audit trail for a business-critical decision looks like.

Regulators are starting to ask those questions. The EU AI Act transparency requirements come into force in August 2026. The Colorado AI Act, which regulates high-risk AI systems, commenced on June 30. Companies that built their agent infrastructure without governance baked in are now retrofitting controls — an expensive and difficult process.

Agent/works is positioned as the alternative: build governance into the architecture from the start, so you never have to bolt it on later.

What This Means for Business

There are two ways to read this announcement depending on where you sit.

If you are a data or engineering leader, Agent/works is an architectural pattern signal. The message from the summit floor this week is consistent: agentic AI is now production infrastructure, and the tooling to run it safely is maturing fast. Governance, evaluation, and cost controls are not nice-to-haves. They are the difference between AI that your CFO can defend and AI that becomes a liability.

If you are a business leader watching from outside the technical detail, the implication is simpler. The era of running AI agents as experiments is over. The infrastructure to run them as accountable business systems is now available. The question for your organisation is whether you are building toward that standard or still running untracked pilots.

The enterprises that are winning with AI in 2026 are the ones where a non-technical business leader can point to their AI systems and answer two questions: what did that agent do, and what did it cost? Agent/works is an attempt to make those questions answerable at scale.

Understanding how agentic systems are structured, governed, and evaluated is now a core skill for both technical and business teams. Enterprise DNA’s learning platform includes structured pathways for teams making exactly this transition — from data literacy into AI-native operations.