Interview with Nir Poleg from Port.io

1. You’ve played a significant role in scaling tech companies through high-growth phases as a substantial part of your career trajectory. As the developer landscape pivots towards agentic development, what is the one nitty-gritty that should be given more focus, especially for businesses at the bleeding edge of the developer-AI transition?

The industry is fixated on the wrong variable. Everyone’s asking which model, which agent, which framework. Those change every week. The thing that deserves more focus is the foundation underneath them, namely how scalable and robust your setup is as the ground keeps moving.

Every wave of engineering, be it cloud, microservices, CI/CD, made teams more powerful and added new complexity. AI is the biggest wave yet, and it’s compounding faster than anything before it. Our customers are operating under real change, pressure, and uncertainty. The problem doesn’t get simpler from here. It gets harder.

So the nitty-gritty is this: don’t optimize for today’s agent. Optimize for the ability to swap it out tomorrow without re-architecting everything. Build for flexibility and durability, not for the tool that’s winning this month. The companies that get this right won’t be the ones with the cleverest agent. They’ll be the ones whose foundation held up over the long term.

2. The agentic promises of massive productivity are often disconnected from the reality of the AI chaos that organizations are facing today. Where do you observe the biggest disconnect between ‘how’ organizations want to adopt AI in their product development lifecycle and the chaotic nature of managing these agents?

The disconnect starts in the boardroom and ends on the engineering floor.

CEOs and boards read that agent-to-human ratios are heading past 100 to 1. They want autonomy, and they want it now. The floor sees something else entirely: agent sprawl. Agents spun up in every corner, no ownership, no visibility, and every so often one takes a destructive action in production. That gap between ambition and reality is real, and most of it is change management, not technology.

But the disconnect people actually miss is where the bottleneck now sits. For years the constraint was writing code. AI is dissolving that. So the constraint moves downstream — to review, testing, governance, incident response, and the plain question of whether you can trust what an agent just did in production. Most organizations are still optimizing the part AI already solved.

MIT found that 95% of enterprise AI pilots returned nothing. Not because the models were bad — because of brittle workflows, missing context, and no standards. The chaos is downstream. Very few people are looking there yet.

3. The current developer arena is heavily defined by ‘AI chaos’ and disconnected promises, as mentioned in the previous question. To what extent is Port’s evolution into an Agentic Engineering Platform a response aimed at stabilizing this chaos for engineering teams?

Entirely, and by design. Port was built to tame engineering chaos: one platform to build, govern, and operate the software lifecycle. Agentic chaos is that same problem, made exponentially worse by AI. Our answer keeps the same shape and extends it to agents — a context lake, guardrails, and a place for humans and agents to work together.

The difference is we can show it, not just say it. When you ground an agent in the context lake, it stops guessing and stops asking. Token spend drops because the agent already knows your world. That’s a measurable outcome, not a slogan.

And we put the proof in the open. A free product. A live demo with no email gate. A public roadmap. Dozens of hands-on workshops. We’d rather show our work than market around it.

We’re not defending a position in an existing market. We’re building the category — the Agentic SDLC platform.

4. Port’s evolution into an Agentic Engineering Platform centers on a context lake and strict guardrails, allowing AI agents to query software catalogs and infrastructure via open standards. Is providing real-time, structured context really the ‘make-or-break’ factor for designing trustworthy AI agents across production?

Yes. Without qualification.

An agent is only as good as two things: what it knows, and what it’s allowed to do. Context is the first half. Guardrails are the second, and they’re not optional — give an agent perfect context and no limits and you’ve just made it faster at doing damage.

Start with context. Drop an agent into your environment blind and it can’t act safely. It doesn’t know this API is PCI-compliant, owned by one team, with three downstream services depending on it. So it guesses. In production, a wrong guess is expensive.

Real-time, structured context changes the question the agent asks. Not “what is this?” but “what is this, who owns it, what’s the blast radius, and what am I allowed to do here?” Guardrails answer that last part — enforced in the platform, not left to the agent’s judgment. Context tells it what’s true; guardrails tell it what’s permitted.

That combination is the make-or-break. It’s the line between an agent that looks impressive in a demo and one you’d trust on-call at 2 a.m.

5. One of the strategic pillars of Port’s new agentic vision is a dedicated human-to-agent collaboration interface, helping on-call teams manage autonomous tasks. In an industry obsessed with total automation, how did you land on “human-in-the-loop” as a core strategic bet for Port’s product direction?

We didn’t choose human-in-the-loop as a compromise. We chose it by watching where agents get into trouble.

“Agent” is not a magic word. It’s not a silver bullet. An agent with autonomy and no boundaries isn’t productivity. It’s a liability waiting to surface. The real choice was never full automation versus full manual control. It’s autonomy inside clear definitions, governance, and guardrails, with a human at the decisions that actually matter.

So we draw the line by consequence. Routine, reversible, low-blast-radius work? Let the agent run. Irreversible, high-blast-radius, or ambiguous work is different. A human reviews and approves. When the predicted blast radius is too large, you want a person in the room. That’s not fear of automation. That’s how you earn the right to scale automation even more.

It’s also why we built a collaboration interface, not just an automation engine. On-call teams need to see what agents are doing, step in, approve, or stop them.

6. As VP of Strategic Initiatives, your focus is on driving high-impact decisions that dictate Port’s wins in an already crowded industry- one that’s also rapidly evolving. With LLMs and infrastructure frameworks changing weekly, what’s your baseline for distinguishing between a short-lived industry trend and a long-term strategic bet that Port must own?

My baseline is the pattern. The way we build software keeps changing. Every wave makes engineering more powerful. Every wave also brings new chaos. That pattern has never broken.

So I don’t bet on the wave. I bet on the constant. Which model, which agent, which framework leads this quarter — those are trends. The chaos every wave leaves behind is the long-term bet, because it never goes away. That’s what Port exists to solve.

And the direction isn’t subtle: more agents, everywhere, across the whole lifecycle. That doesn’t shrink the chaos, it actually multiplies it. So we don’t try to pick the winning agent. We stay open, bring your own, and build the layer every agent needs no matter which one you run: one platform to build, govern, and operate the SDLC, so every team can deliver as fast as technology allows.

My prediction: The AI honeymoon is over. 2026 and 2027 won’t be won by whoever has the flashiest agent. They’ll be won by whoever controls the chaos those agents create. That’s the bet we own.