AI coding agents are blowing through budgets — Replit, Kilo Code, and Symbotic explain how they're managing it

At Kilo Code, engineers now read or write code themselves only about 1 percent of the time, according to co-founder Emilie Schario; agents handle the rest. That shift sits at the center of how tech leads from Replit, Kilo Code, and Symbotic are talking about AI coding agents that are blowing through budgets, as covered by VentureBeat. The story is less about whether agents can write code and more about what happens when almost all of the work moves off human keyboards and onto systems that burn tokens at scale.

The operational questions are concrete. Teams have to decide which systems are safe to hand over to agents, who cleans up when models goof up, and how to support multi-model architectures instead of a single default model. Those choices are not abstract product strategy; they sit inside day-to-day engineering workflow, review paths, and the cost of every run. Token usage becomes a first-class systems concern alongside correctness and ownership of the resulting code.

For engineers and builders, the 1 percent figure at Kilo Code makes the handoff problem personal. If humans barely touch the code, the remaining human work concentrates on oversight, integration, and recovery when an agent fails. That changes who needs tooling, who owns failure modes, and how review and incident response should be designed when the primary author is an agent rather than a teammate.

The market context is the tension between rising agent use and rising bills. Replit, Kilo Code, and Symbotic are treating budget pressure as something that has to be managed, not as a temporary spike. The open question those leads put on the table is whether skyrocketing token spend maps to real progress in delivery, or whether it is mostly burned IT budget with thin returns. That framing turns cost into a competitive and operational signal, not just a finance line item.

The practical takeaway is to treat agent adoption as a systems and cost-control problem as much as a productivity one. Watch how teams set boundaries on which systems agents may touch, how they assign cleanup responsibility after model mistakes, and whether multi-model setups are used to control quality and spend rather than only to chase capability. Until those controls are explicit, high agent utilization can look like progress while the budget is doing most of the talking.