Exclusive: Deliveroo software engineers hand coding to AI agents
Deliveroo software engineers are almost fully letting AI agents write their code, with one employee telling City AM they have barely coded manually for almost a year as the food delivery giant embraces ‘agentic’ workflows.
The London-listed firm is among a growing number of tech firms rapidly evolving software engineering roles around AI, as developers shift from writing code themselves, to supervising fleets of AI tools.
A Deliveroo software engineer told City AM: “I now orchestrate a little team of AIs, rather than writing manual code anymore”.
The employee said they now begin tasks by pasting work tickets directly into AI systems connected to internal codebases.
“It can gather additional information for itself as well”, they said. “My workflow is very much approving the actions that the agent takes”.
The engineer said roughly “90 per cent” of their work now involved overseeing AI systems, with manual intervention only required when the models get confused or lack company-specific knowledge.
“Once I’ve got it out of the ditch it’s got stuck in, then it tends to run on its own”, they added.
A Deliveroo spokesperson said: “AI engineering and productivity tools are already helping us accelerate innovation and raise the bar on the quality of service we deliver for local businesses, customers and riders.”
“Our focus is on empowering our teams, and creating an environment where our engineers can leverage the most powerful tools to deliver real impact and develop their own abilities for the future of engineering.”
Deliveroo ramps up AI push
City AM understands Deliveroo actively encourages engineers to use AI tools as part of its development process, though the firm does not mandate specific workflows or require staff to rely on particular models.
The engineer said AI adoption is “very strongly encouraged” internally, with dashboards tracking how frequently tools are used across teams.
“There is an internal dashboard that tracks people’s AI usage”, they said, adding that there was internal speculation that it could eventually be factored into staff performance reviews.
City AM understands Deliveroo measures staff based on overall impact and efficiency.
The company is also understood to dispute any claims of reducing engineering headcount, despite fears across the industry that AI coding tools could wipe out junior developer roles.
Still, the employee acknowledged hiring patterns linked to AI appear to be changing across the field: “I know engineering managers at other companies and they’ve been explicitly told don’t backfill junior to mid-level roles”.
Major tech firms are racing to embed AI agents into software development across the board.
Google recently said 75 per cent of all new code at the company is now AI-generated and approved by engineers, up from 50 per cent last year.
Spotify and Anthropic have also disclosed extensive AI-generated coding, while Microsoft executives have warned firms risk “hollowing out” the next generation of software engineers if junior hiring slows too aggressively.
Rise of the AI manager
The Deliveroo employee said the role of software engineers is increasingly becoming one of supervision and orchestration.
“The skill that really matters is getting the right context into the agent”, they said.
This could create new risks around oversight and quality control, as firms increasingly hand responsibility to autonomous systems.
“There are still human-gated steps”, they added, “you just don’t want to give an agent free rein to push code into the app”.
But City AM understand some firms are already moving closer to fully automated loops.
A software engineer at US delivery giant Doordash, which acquired Deliveroo in May 2025, told City AM they use AI agents to write code while their manager uses separate AI agents to review it, leaving minimal human involvement in the process.
The rapid adoption of coding tools has also triggered mounting concerns over cost and productivity, with some firms reportedly racking up six-figure monthly AI bills as engineers run multiple coding agents simultaneously.
Research has meanwhile cast doubts on whether AI tools always indeed improve output, as some studies suggest various systems generate large amounts of low quality ‘workslop’ that later needs fixing by humans.
Even so, firms appear determined to push ahead. “We’re given the perspective of – don’t worry about cost, just burn all the tokens you want”, said the employee.