Editorial illustration for Kilo Code Engineers Write Code Just 1% of Time as AI Agents Dominate
Engineers Write Just 1% of Code as AI Agents Take Over
Emilie Schario has a number for you: 1%. That's how often engineers at Kilo Code, the AI coding startup she co-founded, actually read or write code themselves anymore. The other 99% belongs to agents.
Schario laid out the shift at VB Transform 2026, and it wasn't framed as a warning. She called it the new normal.
That normal comes with bills attached. Token costs are climbing fast enough that dev teams now have to justify every agentic workflow against what it's actually producing, not just what it promises. Jared Go, distinguished engineer for AI and cloud at Symbotic, argues the answer isn't pulling back, it's getting sharper about what to hand over.
He's found agents excel at greenfield work, building fresh codebases with no legacy baggage. Brownfield work, patching and maintaining what already exists, remains the harder problem, one that still pulls humans back into the loop for judgment calls agents can't make on their own.
Replit's take on managing that tension, and the ballooning budgets that come with it, looks different.
At Kilo Code, engineers are reading or writing code themselves only about 1% of the time now, according to co-founder Emilie Schario — the rest is agents. That shift is forcing new questions onto dev teams: which systems are safe to hand over, who cleans up when models goof up, how to support multi-model architectures, and whether skyrocketing token bills mean real progress or just burned IT budget.
Why this matters
Schario's 1% figure is a useful gut check, not a milestone to applaud. If engineers at Kilo Code spend nearly all their time reviewing, prompting, and cleaning up after agents rather than writing code, the job has already changed shape, whether or not anyone's updated the job description. For founders watching token bills climb, the harder question isn't whether agents can write code, it's whether the output justifies the spend once you factor in the human time spent supervising it.
Replit and Symbotic facing the same tradeoffs suggests this isn't one company's quirky workflow, it's becoming the default operating mode for teams that bet early on agents. That raises real accountability questions: who signs off on what an agent touches, who's responsible when a model breaks production, and how you staff for a world where "writing code" is a small fraction of an engineer's day. We'd treat this as an early signal worth tracking closely, not a victory lap.
The teams that figure out governance around agent work, rather than just adoption speed, are the ones worth watching next.
Related Reading
Business & Startups Nordic pilot adds Gemini for Education, NotebookLM to boost AI literacy
Business & Startups Kling launches Video O1, all-in-one model with MVL bridge using transformer
Business & Startups DeepSeek Seeks More Capital Weeks After USD 7B Funding Round
From Archives Palantir CEO Alex Karp calls AI industry 'Marxist' after strong quarter
From Archives AWS to sell enterprise security controls for startup Superblocks
Common Questions Answered
What percentage of time do engineers at Kilo Code spend actually writing or reading code?
According to co-founder Emilie Schario, engineers at Kilo Code now spend only about 1% of their time reading or writing code themselves, with AI agents handling the remaining 99% of coding tasks. This dramatic shift was presented at VB Transform 2026 as the new normal for development teams adopting agentic workflows.
What are the main challenges that rising token costs are creating for dev teams using AI agents?
Rising token costs are forcing development teams to justify every agentic workflow against its actual output and productivity gains rather than just implementing agents without cost consideration. Teams must now carefully evaluate whether the expense of running AI agents is justified by the results they produce.
How has the role of engineers changed at Kilo Code with the dominance of AI agents?
Rather than writing code directly, engineers at Kilo Code now spend most of their time reviewing agent outputs, prompting AI systems, and cleaning up errors made by models. This represents a fundamental shift in job responsibilities where supervision and quality control have become the primary engineering activities instead of code creation.
What key questions are dev teams now facing regarding AI agent implementation?
Development teams are grappling with critical questions including which systems are safe to hand over to agents, who is responsible for cleaning up when models make mistakes, how to support multi-model architectures, and whether skyrocketing token bills represent genuine progress or just wasted IT budget. These questions reflect the operational and financial complexities of scaling agentic workflows.
Further Reading
- How 7 Kilo Code Engineers Run Up to 20 Parallel Agents and Still Stay Sane - Kilo Blog
- Kilo Code: An open source AI coding agent that works with multi-file changes, debugging, docs, and terminals - Tessl
- How to Reduce AI Coding Costs for Engineering Teams - Kilo
- Beyond Autocomplete: Best Agentic Coding Workflow in 2026 - Kilo
- AI Coding Assistant Buyer's Guide 2026 - Kilo
