Obsolete Yourself
Treat every repeatable thing you do as something to replace with software or an agent.
- Difficulty
- Moderate
- Time to result
- ~weeks to results
- Steps
- 4
- Confidence
- 90%
A working definition of a great engineer that Singhal has carried across his whole career and now supercharges with AI: the best engineer is someone who obsoletes themselves from everything they do. In the AI era, anyone — not just engineers — can act on this by writing agents to automate the repeatable parts of their own job, freeing time for judgment and the parts only they can do.
Origin
Singhal credits the definition to the best engineer he worked with on his first job, who in turn credited his own father: 'an engineer is someone who obsoletes themselves from everything they do.' Singhal generalized it into a personal operating system for the AI era.
Core principles
- 01If a task is repeatable, it is a candidate for obsolescence.
- 02Obsoleting your own work is not job-elimination — it upgrades the job you're left with.
- 03AI has removed the systems-engineering barrier: you no longer need to be an engineer, only opinionated about what good looks like.
- 04The output shifts from things you ship out the door to internal tools that scale your own judgment.
How to run it
- 1
Inventory your repeatable tasks
List everything you do on a daily or weekly basis that follows a pattern — matching people, tracking a queue, answering recurring questions, compiling status.
Pro tip Start with the least joyful parts of your job; obsoleting drudgery gives the fastest morale and time payoff.
- 2
Pick one and describe the good outcome in plain English
Choose a single repeatable task and tell an AI tool (Claude Code, Codex, Lovable, etc.) what you want, in ordinary language, including what 'good' looks like.
Pro tip You don't need engineering skill — you need to be opinionated about what you want to see and to recognize good when it appears.
- 3
Build the agent and put it into your daily flow
Have the tool build the automation, then actually route the real task through it (e.g., an agent that matches community members to meet, or one that assembles a job-matching mailing list).
Pro tip Standardize on one tool rather than switching constantly — Singhal notes it's hard to context-switch between Claude and Codex, so he standardizes.
- 4
Reinvest the freed time and repeat
Use the time recovered to do higher-judgment work, then ask 'what's the next agent I can build to make my life a little better?' Each cycle deepens the habit until you 'catch the bug.'
Pro tip Keep a human-in-the-loop review where judgment matters — Singhal reads the AI's answers to community questions and notes 'when do the LLMs and I disagree' as his own content edge.
Watch out Some people resist obsoleting themselves because they think a task IS their job. Singhal's counter: obsolete it and you get a better job.
In the wild
With ~125 heads of product in his community, Singhal used to personally think through who should meet whom ('Jay-Z would really appreciate meeting Annie, but I don't know if they've met before'). He replaced this with an agent that does the matching, plus another agent that compiles which of his heads of product are hiring and auto-builds a mailing list of interested candidates.
→ A recurring manual judgment task became an always-on agent — 'the next generation of recruiting' — freeing him from the mechanical part while keeping the value.
Singhal built an AI trained on his content that answers audience questions; he then reviews the questions and answers, spots recurring themes, and writes about where he and the LLMs disagree.
→ He obsoleted the first-pass answering while using the disagreements as the seed for original, differentiated content.
Common mistakes
Guarding a task because it feels like job security
People say 'I don't want to obsolete myself, that's my job.' This preserves drudgery and caps your growth; the tasks worth protecting are judgment, not mechanics. Obsoleting the mechanical work upgrades the role you keep.
Assuming you need to be an engineer to do it
Singhal initially thought this required a systems-engineering skill, then watched his non-engineer wife get huge value. The real requirement is being opinionated about the desired outcome and knowing what good looks like, not coding ability.
Is it for you?
Best for
Product managers and knowledge workers who feel buried in repetitive coordination work and want to reclaim time using AI tools.
Not ideal for
Roles that are overwhelmingly novel, one-off, or relationship-driven with little repeatable structure to automate.
From the transcript
“an engineer is someone who obsoletes themselves from everything they do”
“my stack is what can I do to obsolete anything and everything I do on a daily basis”
“You just have to be opinionated on what you want to see”
“Now, I write software to do that. I write an agent that goes in and actually does matching”
“when do the LLMs and I disagree”
From the episode
Why half of product managers are in trouble
Nikhyl Singhal (Meta, Google)