7 seats left for CTO Circle lunch (Madrid), cognitive debt, and agent harnesses

CTO Circle

Reply-To: marco@pullpo.io

Issue 007

Seventh issue

Cloud and AI infrastructure, built for production

Cloud and AI infrastructure is easier to operate when the platform is clear, automated, and under control.

AstroKube helps engineering teams design, automate, and operate production infrastructure across AWS, GCP, and Azure, combining DevOps automation building blocks with hands-on engineering support.

Start with a free Kubernetes architecture audit. In one week, AstroKube helps you identify bottlenecks, operational risks, and cost leaks in your current setup.

Book your free Kubernetes audit or learn more.

Next community events.

Just 7 seats left - Madrid - Playtomic's office - May 20.

CTO networking, tapas, and short talks from engineering leaders. Apply here!

Short opinionated trend: Hiring devs in the AI era.

A few weeks ago, we hosted a CTO event in Barcelona, and one of the topics that came up was hiring. A few months before that, Ilya, CTO at Factorial, also gave a talk in the community about what they look for when hiring engineers.

My takeaway from both conversations: To design interviews, we first need to reflect on how the role of the developer is changing.

Ticket implementation -> Product engineering

Coding is not enough. This was already true before, but now it matters even more. We need engineers who understand the business, the product, and the users. People who can think, design, and implement end to end.

Why? Because...

IC -> Agent manager

Developers are no longer just individual contributors. They now have access to many agents capable of generating code that would have taken months to write manually.

Their job is becoming more about coordinating them, reviewing them, creating guardrails, giving them the right context, and, most importantly...

Writing code -> Reviewing code

Reviewing. The bottleneck has moved from writing code to reviewing it.

We still need technically strong engineers who can detect mistakes, risks, and opportunities for improvement in AI-generated code. But we also need them to do it while keeping the product, customer, and business context in mind.

I'm not proposing one specific interview strategy. I think many different approaches can work. But whatever process we use, we should make sure we are testing for these shifts.

News, reads and launches.

What I'm Hearing About Cognitive Debt (So Far). Margaret-Anne Storey writes about "cognitive debt": the idea that AI can help teams move faster, but if the code changes faster than the team's shared understanding, the real bottleneck becomes not writing software, but knowing how and why it works.

Open AI Deployment Company and AI services company with Blackstone, Hellman & Friedman and Goldman Sachs. Both OpenAI and Anthropic are now putting serious money behind AI deployment services. Basically helping companies, from large enterprises to mid-sized businesses, move from "AI demos" to real production use cases.

Cloudflare cut 1,100+ people, Coinbase around 700 and Upwork reportedly cut 24% of its team.

If AI Writes Your Code, Why Use Python? Noah Mitchem argues that AI is changing how teams choose programming languages: if agents can now handle the hard parts of Rust, Go, or Swift, then Python and TypeScript lose part of their biggest advantage while faster, stricter languages become much more attractive for production.

deepsec. Vercel open-sourced a security harness that uses Claude and Codex agents to scan large codebases for vulnerabilities, validate findings, assign severity, and turn them into actionable tickets. Basically AI security review running on your own infrastructure.

The Anatomy of an Agent Harness. A harness: the tools, filesystem, memory, sandbox, planning loops, verification, and orchestration that turn raw intelligence into something that can actually do useful work.