Then Google launched Managed Agents in the Gemini API. One API call spins up a full Linux sandbox. The agent reasons, uses tools, executes code — in an isolated, ephemeral environment. Not a demo. Not a chatbot. An employee.
The signal from this week's builder feed is unmistakable: the conversation has moved. "Production agents + guaranteed compute" is the phrase doing the rounds. Not "better chat." Execute.
The typewriter era
Most marketing teams, agencies, and SMBs are still optimising for prompt quality. Better prompts = better outputs. The focus is on writing the right instruction, testing variants, upgrading the model. The mental model is AI as a very fast, very talented employee who responds to instructions.
The content playbook everywhere is: "here is how to write better prompts." The agency pitch is still largely: "we will produce better AI-assisted content than your competitor."
This is the typewriter era thinking. And it is going to look very dated very soon.
What actually changed
The shift that matters is not better prompts. It is agents that operate autonomously. One call to the Gemini API now gives you a full sandbox: reasoning, tool use, code execution. That is not a better typewriter. That is a worker with a desk, a computer, and the authority to act.
Look at what the builders are tracking: Google ships Managed Agents. OpenAI sells Guaranteed Capacity for agents. Anthropic builds Compliance APIs. The infrastructure for autonomous AI operation is being laid, and it is being laid fast.
For a marketing agency or an SMB, the implication is stark. If you are still competing on prompt quality, you are optimising for the thing that stops mattering the moment the agent can operate without a human in the loop.
The companies winning right now are the ones treating AI as infrastructure — as something that executes campaigns, manages workflows, handles routing — not just something that writes better copy.
At Foundry, we have been building toward this. Hermes and the OpenClaw stack are built around the idea that an agent should operate, not just respond. The managed agent model from Google validates the direction. What we are seeing is the same shift: from assistive AI to operational AI.
The businesses that will win the next two years are not the ones with the best prompts. They are the ones with the best agent stacks — the systems that take a goal and execute it without requiring a human to read and approve every step.
The implication
If you are a founder or marketing manager right now, ask yourself: is your AI setup doing the work, or just helping you think of things to say?
If it is the latter, you are still in the typewriter era. The gap between that and what agents can do is going to become visible — to your customers, your competitors, and your P&L.
This week, look at one workflow in your business. Not a creative task. An operational one — reporting, routing, follow-ups, content scheduling. Can you describe it in a way that an AI agent could execute it without you in the loop? If not, that is where the leverage is.