Final week, considered one of our product managers (PMs) constructed and shipped a function. Not spec'd it. Not filed a ticket for it. Constructed it, examined it, and shipped it to manufacturing. In a day.
A number of days earlier, our designer observed that the visible look of our IDE plugins had drifted from the design system. Within the outdated world, that meant screenshots, a JIRA ticket, a dialog to clarify the intent, and a dash slot. As a substitute, he opened an agent, adjusted the format himself, experimented, iterated, and tuned in actual time, then pushed the repair. The individual with the strongest design instinct mounted the design immediately. No translation layer required.
None of that is new in idea. Vibe coding opened the gates of software program creation to tens of millions. That was aspiration. After I shared the info on how our engineers doubled throughput, shifted from coding to validation, introduced design upfront for speedy experimentation, it was nonetheless an engineering story. What modified is that the speculation turned follow. Right here's the way it really performed out.
The bottleneck moved
After we went AI-first in 2025, implementation value collapsed. Brokers took over scaffolding, assessments, and the repetitive glue code that used to eat half the dash. Cycle occasions dropped from weeks to days, from days to hours. Engineers began pondering much less in recordsdata and capabilities and extra in structure, constraints, and execution plans.
However as soon as engineering capability stopped being the bottleneck, we observed one thing: Determination velocity was. All of the coordination mechanisms we'd constructed to guard engineering time (specs, tickets, handoffs, backlog grooming) have been now the slowest a part of the system. We have been optimizing for a constraint that not existed.
What occurs when constructing is cheaper than coordination
We began asking a unique query: What wouldn’t it seem like if the individuals closest to the intent may ship the software program immediately?
PMs already assume in specs. Designers already outline construction, format, and habits. They don't assume in syntax. They assume in outcomes. When the price of turning intent into working software program dropped far sufficient, these roles didn't must "be taught to code." The price of implementation merely fell to their stage.
I requested considered one of our PMs, Dmitry, to explain what modified from his perspective. He instructed me: "Whereas brokers are producing duties in Zenflow, there's a couple of minutes of idle time. Simply useless air. I wished to construct a small sport, one thing to work together with when you wait."
For those who've ever run a product group, you realize this type of thought. It doesn't transfer a KPI. It's unimaginable to justify in a prioritization assembly. It will get deferred without end. However it provides character. It makes the product really feel like somebody cared in regards to the small particulars. These are precisely the issues that get optimized out of each backlog grooming session, and precisely the issues customers keep in mind.
He constructed it in a day.
Prior to now, that concept would have died in a prioritization spreadsheet. Not as a result of it was dangerous, however as a result of the price of implementation made it irrational to pursue. When that value drops to close zero, the calculus adjustments utterly.
Transport turned cheaper than explaining
As extra individuals began constructing immediately, total layers of course of quietly vanished. Fewer tickets. Fewer handoffs. Fewer "are you able to clarify what you imply by…" conversations. Fewer lost-in-translation moments.
For a significant class of duties, it turned quicker to simply construct the factor than to explain what you wished and anticipate another person to construct it. Take into consideration that for a second. Each fashionable software program group is structured across the assumption that implementation is the costly half. When that assumption breaks, the org has to alter with it.
Our designer fixing the plugin UI is an ideal instance. The outdated workflow (screenshot the issue, file a ticket, clarify the hole between intent and implementation, anticipate a dash slot, evaluation the outcome, request changes) existed completely to guard engineering bandwidth. When the individual with the design instinct can act on it immediately, that entire stack disappears. Not as a result of we eradicated course of for its personal sake, however as a result of the method was fixing an issue that not existed.
The compounding impact
Right here's what shocked me most: It compounds.
When PMs construct their very own concepts, their specs get sharper, as a result of they now perceive what the agent must execute properly. Sharper specs produce higher agent output. Higher output means fewer iteration cycles. We're seeing velocity compound week over week, not simply because the fashions improved, however as a result of the individuals utilizing them acquired nearer to the work.
Dmitry put it properly: The suggestions loop between intent and consequence went from weeks to minutes. When you may see the results of your specification instantly, you be taught what precision the system wants, and also you begin offering it instinctively.
There's a second-order impact that's more durable to measure however unimaginable to overlook: Possession. Individuals cease ready. They cease submitting tickets for issues they might simply repair. "Builder" stopped being a job title. It turned the default habits.
What this implies for the trade
A number of the "everybody can code" narrative final 12 months was theoretical, or targeted on solo founders and tiny groups. What we skilled is completely different. We’ve got ~50 engineers working in a fancy brownfield codebase: A number of surfaces and programming languages, enterprise integrations, the total weight of an actual manufacturing system.
I don't assume we're distinctive. I feel we're early. And with every new technology of fashions, the hole between who can construct and who can't is closing quicker than most organizations understand. Each software program firm is about to find that their PMs and designers are sitting on unrealized constructing capability, blocked not by ability, however by the price of implementation. As that value continues to fall, the organizational implications are profound.
We began with an intent to speed up software program engineering. What we're changing into is one thing completely different: An organization the place everybody ships.
Andrew Filev is founder and CEO of Zencoder.
