Shraddha Sunil

engineer @Microsoft. Focused on vertical AI, real outcomes, and industry economics, occasionally angel investing. SF. Down for coffee.

[Shraddha Sunil @shraddhasu]

Agree with you on some of that as well.

Some days I am amazed at how quickly ai fixes something and other days, it creates more problems than solutions.

For now, it’s a bit of both but it’ll be interesting to see how things pan out.

[Shraddha Sunil @shraddhasu]

People always say "the best software engineers will survive" the AI transition.

I used to believe that. Now, I’m not so sure.

I set up Claude on a multi-node cluster to see how it handled a complex, low-level system issue today.

It managed to extract the telemetry, debug a deep distributed systems bug, and figure out the exact fix, entirely autonomously.

If we are already at the point where AI can handle low-level debugging across multiple nodes, the "senior engineers are safe" narrative might have a much shorter shelf life than we think.

[Shraddha Sunil @shraddhasu]

@eduardadotai Maybe browsers don't disappear, they just become infrastructure. Agents becomes the interface

[Shraddha Sunil @shraddhasu]

@craigweiss finally reinforcement learning with positive affirmation

[Shraddha Sunil @shraddhasu]

@peytoncasper The really interesting metric would be output per dollar, not who spends the most on agents.

I suspect the best engineers end up using less compute than you'd expect because they're better at knowing what to delegate

[Shraddha Sunil @shraddhasu]

@sl1ma4 unfortunately this is much less fun than building it

[Shraddha Sunil @shraddhasu]

@yorambac really interesting direction. curious how much capability you found you could preserve through distillation while getting it small enough to run locally

[Shraddha Sunil @shraddhasu]

@vaibhavbetter Robotics probably makes this tradeoff even harsher because the startup needs more capital and more time before the equity gets a chance to become liquid

[Shraddha Sunil @shraddhasu]

@samueljmcd model selection became a personality trait remarkably fast

[Shraddha Sunil @shraddhasu]

@thatguybg The primitives are converging fast. feels like the real differentiation moves to context, permissions, memory and how reliably the agent can actually finish things

[Shraddha Sunil @shraddhasu]

@pcshipp surviving 6 months these days has become a challenge.

[Shraddha Sunil @shraddhasu]

@jiang_kev @polarbrowser Windows being 70% of desktops is one of those stats that makes living in the Bay Area feel like a bubble

[Shraddha Sunil @shraddhasu]

@hnshah unfortunately the company API is undocumented and half the endpoints are people

[Shraddha Sunil @shraddhasu]

@MartinGTobias also explains why fundraising can look bizarrely easy from the outside for some founders.
You're only seeing the last 10% of the process

[Shraddha Sunil @shraddhasu]

@Edybuilds secure until the agent discovers tool use 😉

[Shraddha Sunil @shraddhasu]

@MartinGTobias probably an underrated advantage of building something small first - you learn which problems actually require capital and which ones just require work.

[Shraddha Sunil @shraddhasu]

@Katie_Roof $100m ARR somehow went from IPO territory to “nice start”

[Shraddha Sunil @shraddhasu]

@vineerpasam but it hasnt really been as impactful in other areas of my life yet.

[Shraddha Sunil @shraddhasu]

@vineerpasam Absolutely.
Being able to go from knowing very little about something to a working prototype in a weekend is a pretty meaningful change

[Shraddha Sunil @shraddhasu]

@zuess05 maybe the bigger change is that the downside of trying has collapsed.

You can build something meaningful with 2 people and very little capital, while the upside hasn't really changed