“Vibe coding slop”: Port’s CEO on the problem with ungoverned AI dev
The New Stack Adrian Bridgwater
Port launched an AI Builder tool that puts guardrails on "vibe coding," making AI write plans humans approve before any code ships. The pitch: skipping AI is the real risk now, but skipping oversight is what actually creates the mess.
Based on reporting by The New Stack, Adrian Bridgwater — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Vibe coding has a reputation problem, and Port's CEO Zohar Einy isn't shy about naming it. He calls the unsupervised version "agentic chaos" or, more bluntly, "vibe coding slop" — AI churning out code fast, with nobody checking whether it actually holds together. On Tuesday, Port rolled out AI Builder, a tool meant to fix that by wrapping natural-language development in versioning, audit trails, and mandatory human sign-off before anything gets built.
The mechanism is called Plan Mode. Instead of an agent just generating code on request, it drafts a plan, asks clarifying questions, and sits still until a human approves it. Every plan gets versioned and saved, so there's a paper trail if something breaks six months later. Einy frames this as a leveling force for junior engineers specifically — under Port's system, he says, they get more oversight than senior engineers used to have, which keeps accountability intact even as the coding bar shifts.
And that shift is real. Einy argues the valuable skill now isn't memorizing syntax, it's reading code and understanding design well enough to judge what an agent produced. Port's Context Lake is the piece that's supposed to make that judgment easier: it pulls in an organization's actual stack, team structure, prior architecture decisions, coding patterns, SLAs, and operational data, so the AI isn't building from a generic template but from the specific mess (or order) a company already has. Validation, in this model, splits between human review and AI agents pressure-testing the output — QA stops being a purely manual chore.
Port built the new tool on top of its existing Agentic SDLC Platform, which already handles workflow orchestration and agent management, and it's positioning AI Builder for things like autonomous bug resolution off Jira tickets, regression testing, and tracking AI costs and engineering performance. The company name-checks Claude, Cursor, and Lovable as inspiration, betting that prompt-driven development is creeping out of app code and into platform engineering itself — a trend IDC's Jim Mercer backs, saying the real differentiator soon won't be who can define agentic workflows in plain language, since everyone will, but who has context and governance at scale.
Einy pushes back hard on the idea that speed itself creates technical debt. His line: debt comes from bottlenecks and slow decisions, and humans rack it up when they're overwhelmed, not when they move fast with oversight. Whether that holds up once these context-aware pipelines run at real enterprise scale is the open question — but the direction, at least, is a deliberate retreat from pure vibes toward something more accountable.
My take — AI-written commentary, not fact-checked reporting
I'll believe "AI with oversight removes the bottleneck" when I see a team ship something gnarly through Plan Mode without quietly rubber-stamping every approval to hit a deadline — that's usually where governance theater collapses into the exact slop it was meant to prevent. Still, credit where due: this is a healthier instinct than most of the vibe-coding hype, which treated review as a buzzkill rather than the actual job. Context lakes and audit trails aren't sexy, but they're the difference between an engineering practice and a demo.
Read more about this at: The New Stack