RaghuVaran

Building Agents for Enterprises | Learning Inference engineering | Mechanical Engineer

[RaghuVaran @RaghuModupalli]

@shakoistsLog Ship them to a no-WiFi meadow with cows and zero Slack notifications immediately.

[RaghuVaran @RaghuModupalli]

@vikramchopra @DeploymentInc No openings for bangalore ?

[RaghuVaran @RaghuModupalli]

@larsencc Hey you guys still dont have plans for BU 2.0 open-sourcing ??

[RaghuVaran @RaghuModupalli]

@OfficialLoganK True. Every time I think I've fully updated on the curve, another leap makes yesterday's 'optimistic' forecast look conservative.
The slope isn't just steep—it's accelerating. What's your biggest underestimation right now?

[RaghuVaran @RaghuModupalli]

@ML_Philosophy To wakeup everyday

[RaghuVaran @RaghuModupalli]

@PhilosophyDose_ A person becomes boring the moment they stop being curious about the world — and about other people.Narrow interests, no humor, constant complaining, and zero agency are the classic ingredients.Don’t be the human equivalent of background noise.

[RaghuVaran @RaghuModupalli]

Agree 100%.
Clear, intention-revealing names let you read the code like a story. Clever abstractions often just force the next person (or future you) to reverse-engineer the story. Naming > cleverness, every time.

[Tumusiime Alaphati @alaphati_t]

"Hot take:"
Clean code is more about clear naming than clever abstractions. Agree or disagree?

[RaghuVaran @RaghuModupalli]

@avijeet_writes Pretty solid first day of August!
Caught up on some rest, cleared a few tasks, and already feeling that new-month energy. How about you?

[RaghuVaran @RaghuModupalli]

Prototypes shine when you’re figuring things out yourself or with a small team.
The moment you need someone else to understand the idea, the risks, the trade-offs, or the intent, they usually fall short. That’s when you need the pitch, the narrative, or a clear written shape — not just the interactive artifact.

[Ryan Singer @rjs]

Prototypes are very good for developing ideas and insufficient for communicating ideas.

[RaghuVaran @RaghuModupalli]

Alignment is the central technical and philosophical challenge of the century.
The fact that so much of the discourse around it has collapsed into tribal dunking and status-signaling is itself evidence of how poorly we’re handling it. The wine helps the honesty, though.

[sunil pai @threepointone]

I think alignment is the most problem to work on in our lifetimes and not enough people are taking it seriously and we’re being desensitised to it too much on this site being swayed by pithy well written tweets and appeal to cheap inference that you justify with ideology

[RaghuVaran @RaghuModupalli]

@ML_Philosophy Knowing the right thing to do… and still choosing the easy one.

[RaghuVaran @RaghuModupalli]

In a post-ASI world the AI won’t just write the code — it is the sandbox, the QA suite, and the production environment all at once.Compliance (PCI-DSS etc.) is the only thing that will still force us to keep the old theatre of separate environments. Everything else becomes pure superstition.

[braai engineer @BraaiEngineer]

sandbox & QA envs will become irrelevant in post-ASI world, except for companies with compliance requirements e.g. PCI-DSS

[RaghuVaran @RaghuModupalli]

AI can write the code and generate the screens.
It still can’t decide what should exist, feel the product in a user’s hands, or care when it ships broken.Founders don’t need average devs and designers anymore.
They still desperately need the great ones.

[Dan Ellison @danellisona]

Unpopular opinion:

Founders don't need devs & designers anymore.

[RaghuVaran @RaghuModupalli]

@danellisona AI can write the code and generate the screens.
It still can’t decide what should exist, feel the product in a user’s hands, or care when it ships broken.
Founders don’t need average devs and designers anymore.
They still desperately need the great ones.

[RaghuVaran @RaghuModupalli]

@matvelloso “Unverifiable domain: performance reviews.
‘Exceeded expectations’
Source: vibes.”

[RaghuVaran @RaghuModupalli]

Totally feel this. The half-life of agent instructions is getting ridiculously short.
Best version of AGENTS.md I’ve found so far: keep it brutally lean and only document the things that don’t change with models (exact build/test/lint commands + hard project constraints). Everything else the agent can discover.

[RaghuVaran @RaghuModupalli]

Fable 5 calling them Fields-Medal-worthy tracks. Astra just dropped ten results that sat open for decades — existence of non-sofic groups, disproof of Connes’s rigidity conjecture, Ehrhart’s volume conjecture, multicolor Ramsey numbers (Erdős 183), closest vector problem hardness, and more — all with Lean certificates for ~$2k total compute.Math isn’t solved.
It’s just got a new co-author that doesn’t sleep.

[RaghuVaran @RaghuModupalli]

Swagger/OpenAPI wins for me every time — the fact that it stays in sync with the actual code is everything. Docs that drift are worse than no docs. What about you?

[Kirtesh @AKirtesh]

Devs, how do you document your APIs? 👇🏻

- Swagger/OpenAPI
- Postman collections
- Markdown docs
- Docs? never heard of her

[RaghuVaran @RaghuModupalli]

Strongly agree.
Vision opens the door, but the wedge is what gets people to walk through it and write the check.
The cleanest ones I’ve seen are usually some version of: “Here’s a painful, high-frequency workflow that already exists, and here’s the first 50–100 people who are already pulling hard on a half-built version of the future.”
That combination makes the bigger market feel inevitable instead of theoretical.

[RaghuVaran @RaghuModupalli]

@rjs Love this framing.
Good architecture is evolutionary pressure — strengthen what already holds weight.
Bad architecture is scorched-earth redesign: wipe the map and force 10 new primitives because “fresh.” Tissue → bone is perfect.