“Agents only matter if they can act, not just suggest.”
Redeployed is a weekly newsletter that breaks down one important AI story at a time for leaders in technology. Every issue explains what the shift means for technology companies and how smart leaders can use it to get ahead.
Anthropic just crossed a line most companies weren't fully prepared for.
With the latest updates to Claude Cowork, AI agents can now do more than connect to tools, they can operate them. Open files, navigate browsers, run applications, execute tasks across systems like Slack and Google Workspace, even acting while the user is away, with permission.
This isn't like a typical integration update. It's a shift from AI that helps you work to AI that works on your behalf.
For the past year, most enterprise AI efforts focused on connecting models to tools. You could ask for a summary, draft a document, or retrieve information from internal systems. Now the model doesn't just return an answer. It takes action.
From Tools to Operators
The difference seems subtle, but it changes everything.
Before, AI sat next to your workflow. You prompted it, reviewed the output, and decided what to do next. Now it sits inside the workflow, executing steps, moving across systems, and completing tasks without constant input. That turns AI from a layer into a participant.
Think about what that looks like in practice: a support agent that doesn't just draft a response, but pulls data, updates records, and sends the message. A finance assistant that doesn't just analyze a report, but generates and distributes it. An engineering assistant that doesn't just suggest code, but navigates the repo and applies changes.
The interface is no longer the prompt. It's the outcome.
The Real Shift Is Responsibility
This is where the conversation changes. When AI only suggests, the human owns the decision. When AI acts, ownership gets murky fast.
Who reviewed that change? Who approved that action? Who's accountable when the agent misinterprets context or executes the wrong step? These aren't edge cases. They're the new baseline.
The challenge is no longer whether AI can help teams move faster. It's whether teams can maintain control as AI starts moving on its own. And that's not a tooling problem. It's an organizational one.
What This Means in Practice
As agents move into real workflows, the work doesn't disappear. It shifts. Less time executing repetitive steps. More time defining how those steps should happen, where AI can be trusted, and where it still needs oversight.
Teams are already looking for engineers who can design these systems end to end. Not just connect APIs, but define permissions, build guardrails, and monitor behavior in production. Because once AI can operate in your environment, the risk isn't theoretical anymore. It's operational. And that requires people who understand both the system and its failure modes.
This issue of Redeployed is brought to you by Tecla: As AI agents start executing real work, the risk is no longer theoretical, it's operational. The teams moving fastest aren't replacing engineers. They're adding people who can design guardrails, manage systems, and take ownership in production. Tecla helps companies hire senior tech talent in the U.S. and nearshore who already work alongside AI tools, so teams can scale with control, not just speed.
Automation Without Context Is Fragile
The promise of agents is clear: faster execution, less manual work, continuous operation. But the limitation remains the same. AI doesn't understand intent. It recognizes patterns.
What looks like a routine task to a model might be a critical exception in a real system. What seems like a harmless action might trigger downstream effects no one anticipated. When agents act without context, speed becomes a risk. And without the right oversight, that risk compounds quickly.
Where This Is Headed
Enterprise AI is moving out of the chat window and into the operating layer of software. The companies that win won't be the ones that deploy the most agents. They'll be the ones that design the best systems around them. Clear ownership. Strong guardrails. Observability into what the agent is doing and why.
Because the future isn't AI that answers questions. It's AI that executes work. And the real question is no longer what the model can do. It's who's responsible when it does.
Connect With Other Technology Leaders
If you want to connect with other technology leaders having real conversations about AI and how it is changing business, check out GILD Curated Circuit.
More to come…
Recommended Reads
✔️ Anthropic’s Claude Code and Cowork can control your computer — The Verge
✔️ OpenAI acquires Promptfoo to secure its AI agents — TechCrunch
– Gino Ferrand, Founder @ Tecla