"Nobody writes a line of code anymore" — a four-person AI-native team, with John Wards of Abatable
Six months into leading engineering at carbon-markets startup Abatable, John Wards runs a four-person team where no one writes code by hand. He explains how he shipped a re-platform on a fixed deadline through two departures, why he builds throwaway internal tools in two-day sprints, and how designers now ship the first working version.
- "Nobody writes a line of code anymore. Not one engineer writes a line of code."
- Shipped a re-platform on a fixed deadline — no scope cut, through two engineers leaving — with a team of four.
- "Writing code isn't the slow bit. It's deciding what to do."
- After 25 years of hype cycles: "I should be cynical. I am cynical. This one is different."
When John Wards joined Abatable earlier this year, the carbon-markets startup had four engineers and a tech function that sat inside the product organisation — no separate engineering org, no layers. "My job is to ship product, basically," he told me. Six months in, he leads a team where no one writes code by hand and where internal tools that used to take a quarter get built in an afternoon.
I sat down with John to unpack how that happened so fast. What follows is an edited version of that conversation.
Interview by Samer Bechara.
Landing mid-flight
Wards arrived at an awkward moment: Abatable had won a large piece of work that doubled as an excuse to re-platform, and it came with a fixed external deadline. "You go from a product organisation, where your deadlines are internal, to — oh, we've got to work like an agency, because we've got to deliver something by a deadline," he said. "It's a good muscle to learn. But it's not what the engineers signed up for." He landed with six to eight weeks left on the clock; most of the features were already designed and barrelling toward launch.
His first instinct was to stay out of the way. "My gut working with my boss was: this is running. Am I going to add value, or am I going to get in the way?" Within two or three weeks the answer was obvious. He stepped in — not on the critical build path, but around it. He took project management off the lead engineer's plate, and then took on the grind of bug-fixes and polish that was pulling everyone out of deep work. "I joked, it's like — I'll just sweep up behind you." That let him get up to speed on the codebase while unlocking headspace for the people building the greenfield parts.
The result surprised even him:
Without changing any scope, on time, with two people leaving engineering-wise in those six weeks, and me coming in new — we hit deliverables and the customer was happy. I would have assumed we would not have done that.
He credits the combination of hard-won experience and a tooling shift he'd spent the previous two months living inside.
"Nobody writes a line of code anymore"
Wards spent December and January deep in the new generation of coding models — first Codex, then Claude landing in December. "I'm going, oh my — the things I can personally do with a product-engineering hat on, the game has changed." He walked into Abatable and found a team open to it. The headline result is blunt:
Nobody writes a line of code anymore. Not one engineer writes a line of code.
Getting there was less about the models than about the plumbing. While pushing bug-fix tickets through — "I had five, six terminals running, each ticket, making PRs" — he was quietly rebuilding the team's Git setup and dev tooling to be AI-friendly. The test he applied: a new environment should spin up with one command, for a human onboarding or an agent starting a fresh session. "You don't want too many dependencies. You don't even want to install Docker, because then you've got networking and compliance. Keep it simple — works for an engineer, works for an AI agent."
The disposability of it all is the part that still seems to delight him. To manage his own swarm of agents he wrote a native Kanban-style app to watch the tickets running on his laptop:
I wrote a Swift app that looked like a native app. I've never run a line of Swift in my life. But it works — it looked a bit like Linear, it was really nice. And I don't need it anymore. It's thousands of lines of code I just threw away.
A side effect he didn't expect: the workflow forced the software-development discipline teams always claim to have. "Most teams say they do a proper SDLC, but you don't — in small teams you just Slack someone the design. Now you start forcing it into the ticket, because you start your AI session and say: look at the ticket. So the context has to live in one place."
Two-day swings
Once the delivery crunch was behind them, his boss started testing what the team could really do. One idea: a chat interface over Abatable's data. "I said, yeah — build it. Two days. We built a ChatGPT-type interface with tool calling. You could explore the database, write custom queries. It had evals, it had self-learning, the UI was on brand." They never shipped it — not because it didn't work, but because they hadn't validated the product need. And that, for Wards, is the whole point:
Writing code isn't the slow bit, usually. It's deciding what to do.
The lesson he took wasn't about the feature. It was about the leverage: "Product mind, plus engineering, plus essentially a team of four engineers that is your AI capability — you can knock things out of the park."
Getting designers past Figma
The next challenge from his boss was pointier: how do the designers contribute beyond Figma? Wards didn't want to hand them a terminal. "Putting Claude Code on their machine — that's not going to work." So he built something he calls Studio: a version of Claude Code hosted in the cloud but locked to Abatable's app.
Think of it as Claude cloud, but focused on our app — the only thing it could build was our app.
Under the hood it's a harness wrapping the agent with app-specific prompts and patches, running on fast, disposable VMs from E2B, with a live preview pane so a designer can see the thing they're building. Couple of days to stand up.
The organisational change it unlocked is bigger than the tool. Design review used to be the last step — the ritual of "that's two pixels off, that's the wrong shade of grey" after engineers had already built it. Now it's the first:
Design reviews are done first rather than secondary. The designers build a working version in Studio — sometimes with a stubbed-out backend — and take ownership of the quality. Then the engineers pick it up and make it real.
Engineers, in turn, feed what they learn back into the agent's context files "so it doesn't reinvent a component that already exists — so we're not burning tokens."
A teammate in the channel
The most recent move — barely a week old when we spoke — was putting the agent where the team already works. When Anthropic's Slack "@Claude" tag shipped, his product manager wanted it in their project-kickoff channels. Wards balked at the extra billing line — "I've already paid a subscription for capability" — then realised he'd already built the whole substrate with Studio. Two days later he had his own channel agent: memory that survives the VM shutting down, access to Google Docs via its own email address, Linear, Git, sub-agents, and the ability to kick off a Studio build the team can watch produce a PR.
Then it started being used without him:
An engineer tagged Studio in a thread from a bug last week and said, "Hey Studio, can you research this, and if it's fixable, start a session to fix it." And it went, yep — did it. I didn't ask the engineer to do that. They just knew it was there.
What he'd worry about next
I asked what he'd look at first if he walked into a new company in the position he was in six months ago. His top three were less about technology than about conditions.
Willingness. "I've had to expend no energy convincing anyone of any of this. And I know for a fact that isn't the case in a lot of other jobs." That, he says, would be his biggest worry somewhere new. He's careful to frame it as a cynic, not a booster:
I've been in this industry too long to count — 25 years. I've seen tech come and go, hype cycles. I should be cynical. I am cynical. This one is different.
His conviction has an edge to it: "The job people think they have isn't going to exist for very much longer. A model can write good-enough TypeScript, React, Go — faster, cheaper. There'll be people writing code in specialist areas, but for the web or app developer, the economics have flipped."
Cost. As a startup, Abatable can be flexible on risk. Engineers each get a Claude Max account — "nice and cheap," a couple of hundred dollars — and he doesn't police usage: "I'm not asking them to optimise. I don't care about that. I care about the things we're building." He's clear-eyed about the raw numbers, though. Measured as equivalent API spend, "in June I was nine thousand dollars." In a bigger company without a startup's risk appetite, that flexibility is exactly what he'd expect to have to fight for.
Team shape. When one engineer left, he and his boss decided not to backfill. "We're already going fast. We haven't got enough ideas to demand another engineer." It's not friction-free at four people — "four people book their holidays and there are no engineers" — but it reframes the question of scale entirely:
I can't fathom what a fifty-engineer B2B startup is doing with all those engineers. What are they doing?
Field Notes is a Courtix series of conversations with the engineers and leaders building serious software. If you're a technology leader with a story worth telling, get in touch.