Aaron Levie argues coding is the most viable AI automation target, while Herbie Bradley says it is just the starting point

agree, but the context being “already digitized and sitting in the codebase” is the quiet part. in most knowledge work it’s sitting in one senior person’s head and a shared drive nobody’s cleaned since 2019. the agent’s ceiling isn’t intelligence, it’s how much of that you’ve actually extracted. that’s the work, and almost nobody’s scoping it

@levie Yet its politically incorrect to say AGI is not right around the corner in SV

@levie verifiability cuts the other way here. it's exactly why engineers are still visibly needed. un-verifiable knowledge work is more exposed, not less. no test suite means the slop ships quietly and nobody can prove the human still mattered. the danger is invisible, not absent.

its true but you also have to consider theres a catch up period. The future is unevenly distributed. Anecdotally, many of my junior/mid-level software eng friends are getting laid off. Statistically, this is the first time ever people with a college degree have a higher unemployment rate than people without. There’s nuance but it does seem like the world needs less junior engineers (code monkeys). And if that finds its way to other knowledge work, we need to find other ways to train engineers

@levie My bias: coding is the friendly case because failure shows up in tests, PRs, and broken screens. In business work the hard part is the messy exception: which human owns it, and what is the next safe action?

@levie The 80% stat already shows the shift: human role flipped from writing to overseeing. Overseeing code that ships 8x faster probably needs fewer people than writing it did. Demand looks healthy because 2025 headcount decisions haven't unwound yet. That's the lag, not the floor.

@levie wait, how is it the pinnacle? there's so much beyond code, code automation is just the substrate

@levie At least for the next few years.

As memory-related technologies improve and reach their inflection point, the question of job security will arise again

@jatingargiitk Tell that to the judge :-)

@levie The most interesting “unknown” is how it will impact physical world even before robots become ubiquitous. All human activities are based on knowledge and experience and we may see “sci fi”-like future where everyone has any knowledge they desire at will, but not the experience.

@levie Built an OS in rust recently. Sonnet 3.5 handled a lot but couldn't fix its own bugs. Oversight still essential. Gap closes every release though. People don't go away, the on-ramp just narrows.

@levie Agree. No doubt that we need a ton of people to work on this project of continuous improvement with AI across all spectrums of business. There's so much work to do.

@levie What I think is the biggest risk for coders and other knowledge workers is saying: "do the best for x" and then trust it blindly. Models aren't there yet.

@levie people keep posting this like it's bullish lol. coding is the easiest possible case for agents and we STILL can't take the human out of "wait should we build this." now go try that on literally any other job. yeah

@levie the 'verifiable' point is carrying a lot here. you can verify an app loads, way harder to verify the code holds up in 6 months. that's most of why engineers stick around, not in spite of it

@levie Yep.

@levie Do you think engineers will be overseeing the actual code that is written in six months?

How would that change your thesis?

@levie It's also just outright wrong at times. Add this line, when really you needed a whole separate config block far above for priority execution.

@levie This is the hard ceiling people skip. If coding still needs engineers around the loop, most other AI workflows need clearer judgment, evidence, and rollback surfaces before they can run unattended.