Agentic Engineering is a disciplined approach to AI-assisted software development that emphasizes human oversight and engineering rigor, distinguishing it fr...

Addy's platform is a resource for web developers and performance enthusiasts, offering insights into web performance optimization, browser rendering, and front-end development techniques. Through articles, tools, and case studies, Addy offers insights into measuring and improving web page speed and user experience. Developers can learn about performance budgets, lazy loading, and optimizing web assets to create fast and responsive websites.

Addy Osmani

Agentic engineering is a disciplined approach to AI-assisted software development that distinguishes itself from "vibe coding" through human oversight and engineering rigor. While vibe coding means accepting AI output without review (useful for prototypes and MVPs), agentic engineering involves treating AI agents as tools that handle implementation under careful human direction. The workflow requires writing specs before prompting, reviewing every diff, running comprehensive test suites, and maintaining ownership of the codebase. This approach disproportionately benefits senior engineers with strong fundamentals, as it trades typing time for review time and demands architectural thinking over raw code generation. The rise of AI coding raises rather than lowers the bar for software engineering craft.

Agentic Engineering

We need a better term for the professional version

What agentic engineering looks like in practice, perhaps.

Nice distinction between vibe coding and agentic engineering.
My daily development setup:
<ol>
<li>Define ‘<a href="http://instructions.md" target="_blank" rel="noopener nofollow">instructions.md</a>’ with coding style, project structure, things not to do (e.g do not use any, stricter typechecks), project flow, major components/services. This is one time setup but reduces lot if cognitive load and ‘trial and error’ prompts.</li>
<li>For the new feature, ask gpt to create spec file.</li>
<li>Manually review and edit the spec with clear input, output, some edge cases, the actual process broken down into multiple tasks. All of these with a status field.</li>
<li>Ask clause to create sanity and basic tests cases against the spec. Acts as guardrails.</li>
<li>Ask claude to pick the 1st task from spec based on status, start working on it.</li>
<li>Spawn new agent to run qa (peer review) on the generated code. Based on feedback, ask 1st agent to fix them and keep close eye on the spec and tests.</li>
<li>Once QA and local tests say all good, I manually test it against spec and end goal.</li>
<li>After this, i manually review all the changeset</li>
<li>If something is missed, i update it in spec file and ask claude to re-reciew it and fix it.</li>
<li>Finally run another claude agent to do critical code review from senior engineer perspective to check for severities, owasp issues, backdoor loopholes.</li>
</ol>
I know this is a lot but, i am way more confident with this approach where I still feel accountable for owning the codebase.
Turns out, i was doing agentic engineering and downselling it as vibe coding.

I miss the good old days where Computer Science was not fractured in empty buzzwords.

Addy always brings gems with his writtings

need strong fundamentals for agentic engineering. So software engineers are still valuable.