Vaibhav Gupta

I don’t enjoy the syntax and tooling we use to building AI pipelines. Langchain/ai sdk/et all have never really felt “right”. No type safety, single language support, prompt is hidden until I buy some observability suite. So we made a thing youtu.be/wD3zieaV0Yc?si…

@aaronvi @anshnanda I just wanna code more and tweet less T_T

@anshnanda @vaibcode we can basically just tweet things like this

The hardest thing to build today is an Agent Harness. You can't just vibecode a good harness. A good harness requires care. We think we built the best general agent harness over the past 6 months. When building a Harness you need to consider the following: - Streaming, Tool Calling, Compaction, Memory - Subagents, Background tasks, Steering, Queuing, Monitoring - Persistence, Permissions, Connections, Crons, Workflows Making all of these GREAT, while also allowing the user to customize and share their @chorus_agent is probably the hardest thing we've done as a team. But we believe that the people who build the most useful agents won't even know what a harness is... They will be people who are experts in a given domain. We will empower those people.

@levelsio @dexhorthy and @0xblacklight have a bomb office

Any SF startup office we can work from today? SF cafes are absolutely unworkable, reminds me of Lisbon, no laptop culture which is of course ironic

@tomgreenwald oh yea i'm about to try that soon! we have an agent trying to write BAML pretty much 24/7, so will try and swap to this.

@vaibcode Let's see how well Magnitude can write BAML

Introducing Magnitude. It's a coding agent that runs entirely on open models. It costs 60% less than Claude Code with no drop in performance. Try it now: npm i -g @magnitudedev/cli Here's how it works 👇

Someone really wants our promptfiddle.com api keys... here let me make your life easier. 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 Meanwhile, promptfiddle is now BYOK :(

I’ve given opus enough love. Time to switch to codex.

STOP WRITING YOUR SPECS BY HAND yeah, ai can write your code, but it can also write your specs and PRDs but the same rule as always: slop in → slop out if you outsource the thinking, you’re gonna get garbage lucky you, me and @vaibcode went deep on some powerful techniques for getting really good product specs from coding agents, without slaving over markdown files for hours

@dexhorthy The way we share information between teammates is going to fundamentally change over the next year

@MichaelThiessen html? or was this post slightly sarcasm xD

I’m so sick of Markdown right now. It’s amazing for humans and LLMs, but impossible to process or manipulate otherwise. Is there anything that renders like Markdown (for humans) but has a strict structure (like XML/YAML/etc) so it’s easy to manipulate? Ideally, a superset of Markdown, so it “just works” and renders like normal Markdown. But with added structure that makes it easy to parse and verify and manipulate. I’m building my own thing currently but I’d rather not.

@antoniosarosi @boundaryML “Supposed to” being the key word 😭

Interning at a YC startup > Aggressively stares at Claude in the pool > "When I started working at @boundaryML I noticed that nobody really reviewed code" Bruh we're supposed to be building compilers here 😭

post training is so odd. claude seems to have learned max-line-length...

@lenadroid @aiDotEngineer @WeAreDevs see you there!! :D

So excited to be speaking at AI Engineer World's Fair @aiDotEngineer in San Francisco very soon ❤️‍🔥 And super thrilled to be giving a keynote at @WeAreDevs in Berlin right after 🔥 Looking forward to seeing everyone! Who is coming?

code can be slop, writing cant

At HumanLayer, we’re on a mission to solve the AI slop code problem. In 2025 we open-sourced our Research, Plan, Implement framework, now deployed inside fortune 500s like Block and Uber - places where shipping slop is just not an option And that was just the beginning. Today, we’re opening access to HumanLayer - an Agentic IDE, collaboration platform, and building blocks for your software factory. HumanLayer enables engineers solving hard problems in complex codebases to: > move 2-3x faster across the entire SDLC (not just coding) > maintain rigorous standards for system architecture and program design Hundreds of engineers at companies of all sizes are already using HumanLayer to ship fast without sacrificing quality. I'm excited to invite you to try humanlayer today at humanlayer.com, and I'm even more excited to see what you build. @0xblacklight and I are deeply grateful to our team, our customers who give us so much incredible energy and feedback, our investors who have always been in our corner, and our friends and family who have supported us along this crazy journey if you're a staff or principal engineer trying to make AI coding work at scale for your team, we'd love to hear from you as @swyx likes to say - let's make this the year of no more slop

@dexhorthy Amazing to see it finally public. Been great to use!

@georgecurtiss @PlanetScale @turbopuffer @helixdb @duckdb Why duckdb over clickhouse?

what your database stack should be in 2026: Relational - @PlanetScale Vector - @turbopuffer Graph - @helixdb Analytics - @duckdb

@pat_erichsen @dexhorthy Appreciate the support 🐑🐑 What should we talk about next?

always enjoy `ai that works` w/ @dexhorthy & @vaibcode great vibes, feels like a lunch and learn at the office w/ your two sharpest coworkers

@georgecurtiss oh i agree, but alas not everyone acts as such :')

@vaibcode I dont want them to be nice. I want them to be fucking real. If I go on a date, I don't want some girl to say "this was nice, lets just be friends" 😂 Im a fucking adult, say to my face you're not interested and then ill go find someone else who is.

“I’m not ready for a relationship” The investors are just doing their job and being nice. They dont have to pinpoint it, but basically first date didn’t feel right. And that’s ok. it did suck in the beginning to hear all these lame excuses from vcs 😭