In recent days several prominent developers have described a shift in how they work: OpenClaw's Peter and Claude Code's Boris said they no longer focus on hand‑crafting prompts. Instead, they design loops — control cycles that generate and refine prompts for agents. At Google, Addy Osmani labeled this approach "Loop Engineering," presenting it as the fourth paradigm after Prompt, Context and Harness Engineering.
Prompt Engineering, Context Engineering and Harness Engineering have each emerged over roughly the past two years and became essential skills in turn. The new term signals another change in emphasis: where effective prompt writing, context management and integration harnesses were once the priorities, attention is now moving toward designing cyclic, self‑refining control mechanisms.
In practice: short skill half‑lives
A recurring theme in the discussion is the rapid obsolescence of practitioner skills: techniques that developers spend months mastering can be partially or wholly superseded within weeks by a new model or approach. The notion of a "half‑life" is used to describe how quickly the utility of a specific technique or role diminishes when successive paradigms arrive.
Concretely, Prompt Engineering produced courses and job titles, then Context Engineering shifted priorities, Harness emerged before many had fully adopted the prior approach, and now Loop has arrived. The result is that professional development must keep pace with technological change at an accelerating rate: the skills people grind to learn today can become marginal faster than they can be absorbed.
Why this matters
Rapid paradigm shifts raise structural questions for technology workforces and training programs. If investing time in a skill risks being made obsolete quickly, rational choices about career development may change: some argue it can be sensible to delay deep investment and instead wait for models that internalize those techniques.
That argument is not definitive. Practical roles, processes and tools still bridge paradigms, and hands‑on expertise remains valuable for many teams. Nevertheless, the debate highlights a tension between the pace of technological progress in AI and the time required for people to learn and professionalize skills.
Conclusion
Calling the newest approach "Loop Engineering" reflects a renewed focus on cyclic, self‑regulating control systems. The rapid succession of engineering paradigms forces companies and practitioners to reassess how to allocate effort between transient, tactical skills and longer‑term capabilities, a question that will remain central as models and workflows continue to evolve.
