In a provocative analogy that cuts through the hype, developer Javier Salcedo Puyo compares AI coding agents to anabolic steroids. Published on DEV.to on October 4, 2026, the piece argues that while these tools allow for massive gains in software quantity, they carry significant risks to individual skill retention and professional independence. Salcedo Puyo posits that agents enable uninitiated users to achieve results previously requiring years of effort, but warns that this convenience comes at the cost of deep understanding and long-term career resilience.

The Bodybuilding Eras of Software

Salcedo Puyo maps the history of programming to the eras of bodybuilding, from the experimental 'Bronze Era' of COBOL to C, to the 'Silver Era' of C++ and Java. He describes the current 'Convenience Era'—characterized by JavaScript, Electron, and AI autocomplete—as the 'Golden Era' of bodybuilding, where production speed increased but quality often plummeted. This period, which he dubs the era of 'human slop,' saw performance and correctness sacrificed at the altar of developer speed, much like the early adoption of Performance Enhancing Drugs (PEDs) that prioritized size over health. The argument culminates in the 'Agentic Era,' which Salcedo Puyo likens to the 'Mass Monster Era' of the 1990s. In this phase, software production speed becomes the primary goal at all costs. He notes that while domain experts can use agents to amplify their work by orders of magnitude, the average user may simply be producing bloated, unverified code. The comparison highlights that while professionals can tolerate higher workloads and recover faster, the long-term structural integrity of the 'body'—the codebase and the developer's skills—remains under threat.

Risks of the PED Stack

The core dangers identified are skill atrophy, economic dependency, and quality degradation. Salcedo Puyo warns that relying on agents may cause developers to lose the patience and capacity to navigate unfamiliar codebases. He describes this as a form of 'tool dependency,' where a developer becomes unable to function without their AI subscription. This mirrors the physical atrophy seen in bodybuilders who stop training while maintaining their drug regimen, leading to a fragile professional state that collapses without constant external support. Economic dependency is another critical flaw. Unlike local tools, agents require ongoing subscriptions, tying a developer's productivity to a vendor's pricing and availability. Salcedo Puyo points out that current token prices are heavily subsidized, creating a false sense of viability. He argues that for most users, the cost of maintaining agent-generated code—through tokens, money, and energy—will eventually exceed the cost of manual verification. This creates a scenario where the 'free cycle' offered by the industry is actually a trap that drains resources and limits autonomy.

Who Should Take the Cycle?

Salcedo Puyo’s verdict is not a total ban, but a strict gatekeeping of usage. He advises that amateurs and novices should avoid agents until they have reached their 'natural limit,' much like athletes who must build a baseline of strength before introducing PEDs. For juniors, the risk is becoming a 'wimp' who cannot put in the work, while veterans with decades of experience can use agents to streamline bureaucratic tasks and focus on complex problem-solving. He cautions that the 'Dunning-Krueger effect' will hit hard, with overconfident noobies believing agents are magic genies rather than tools requiring supervision.

Key Takeaways

  • Coding agents are compared to PEDs: they boost output but carry risks of skill atrophy and dependency.
  • Novices should reach their 'natural limit' before using agents to avoid becoming tool-dependent.
  • Current token pricing is subsidized and may not reflect the true long-term economic cost of agent usage.
  • Veterans can use agents to automate tedious tasks, but must remain responsible for code quality and verification.

The Bottom Line

Use AI agents as a force multiplier for your existing expertise, not a replacement for it. If you haven't hit your natural limit yet, skip the cycle and build the muscle first.