Boris Tane, founder of the Cloudflare-acquired observability startup Baselime and current founder of nominal.dev, published an essay on March 14, 2026 coining the term "slop creep" to describe the slow, invisible degradation of codebases through an accumulation of individually reasonable but collectively destructive decisions made by AI coding agents. Writing from personal experience building Baselime, Tane argues that mistakes which once took months of compounding human error to reach crisis point now arrive in days when a coding agent is in the loop. Claude Code serves as his primary example of the category of tool driving the problem.
The central technical argument is that coding agents lack holistic system understanding — they respond to prompts and see only the files they are given, with no memory of how a system evolved and no foresight about where it needs to go. Tane frames this as the removal of a natural circuit breaker that previously existed in software teams: a junior engineer with poor instincts was self-limiting because slow human execution gave teams time to notice and intervene. Agents eliminate that speed limit entirely, enabling indefinite accumulation of subtly wrong abstractions while <a href="/news/2026-03-15-codegen-is-not-productivity-wrong-metric">surface-level productivity metrics stay green</a>. When the reckoning arrives, Tane writes, it lands as a single expensive rewrite rather than a series of manageable course corrections.
Tane's proposed remedy is not abandoning agents but radically reforming the planning phase. He advocates a <a href="/news/2026-03-15-minimap-local-ui-for-git-native-roadmap-files-in-human-agent-workflows">research-plan-implement iterative workflow</a> in which engineers explicitly define key abstractions, data models, and service boundaries before handing work off to an agent, treating the agent's first draft as a starting point to tear apart and annotate rather than a finished plan. The engineer's job shifts from typing to thinking, with the agent executing within human-defined constraints rather than walking through one-way architectural doors unilaterally. He is explicit that schema decisions, service boundaries, and key abstractions should never be the agent's call alone.
The essay carries a notable commercial dimension worth disclosing. Tane published it exactly two weeks after announcing nominal.dev, a platform building AI agents for autonomous production incident response, with early access rolling out April 2026. The piece explicitly names observability gaps as a recurring failure mode he attributes to unconstrained agents — the precise category nominal.dev is built to address. Tane's career arc from Baselime through Cloudflare's Workers observability team to nominal.dev forms a coherent through-line, and the slop creep essay functions as credible category-defining thought leadership that creates urgency for the product without directly selling it. The unresolved irony is that nominal.dev deploys the same class of autonomous agents Tane warns about, operating in production environments at the highest-stakes moments — a tension the essay does not address.