Pranshu Khanna

🎤 Sasha Denisov (CTO, brainform ai; GDE) on shipping AI on-device: model choices, runtimes, and where CPU, GPU and NPU behave differently. 🎤 Jacek Golebiowski (Co-founder & CTO, @distil_labs) on post-training small language models using your production traces & a live build.

thu 15 oct, 18:00–21:00 distil labs, Schönhauser Allee 10-11, Berlin thanks to everyone who's signed up. once capacity is reached, new registrations go on the waitlist. if you've been meaning to join, register now: luma.com/3q4cjah8

sending the same agent step to a frontier model thousands of times a day? we're hosting a models night on replacing those calls with small models: post-training, evals, a live build + on-device AI Oct 15, 18:00–21:00 · distil labs join us luma.com/3q4cjah8

1 Sept

love open source programs that show up where maintainers are

Santosh Yadav | GDE 🅰️ | GitHub ⭐ | MVP@santoshyadav.dev

CodeRabbit is free for Open Source, and with out 10 million commitment it shows that we are serious about supporting OSS maintainers. The below data just proves the belief OSS community has, 195% increase Since Jan 26, half a million reviews done in only July.

27 Aug

are we back chat? someone just launched twitter.now after X dropped the brand. claim is basically: → Twitter is reclaimed → not X → early access → "freedom of speech, not reach" → trust signals no idea if this is real or just an internet moment.

pran.shAwesome Developer Ambassador ProgramsA curated list of developer ambassador, champion, expert, campus, and community programs for builders who write, speak, teach, organize, and create in public.

25 Aug

i made a directory of 96 developer ambassador/champion/campus programs. ♥️ ai, devtools, cloud, OSS, data, frontend, student programs, all in one place. check it out if you write tutorials, build demos, run meetups, contribute, or help devs learn tools. pran.sh/blog/awesom...

github.comGitHub - pransh15/DeveloperAmbassadorPrograms: A list of Developer Ambassador ProgramsA list of Developer Ambassador Programs. Contribute to pransh15/DeveloperAmbassadorPrograms development by creating an account on GitHub.

25 Aug

i kept losing track of which companies run developer ambassador programs, so I started a list. 📜 it's grown into 90+ programs - Microsoft, Google, NVIDIA, GitHub, HashiCorp, and more. open source and community-maintained, issues/PRs welcome. ♥️ github.com/pransh15/De...

pran.shHighest AI benchmarks ≠ adoption stonksAnthropic's most expensive model is stalling at 11% of spend while cheaper models handle the daily workload. The FT piece, and the Hacker News thread under it, say a lot about what developers actually adopt versus what they admire.

24 Aug

i wrote a little something about adoption stonks for models 📈 also, why highest ai benchmarks are great for investors and headlines but devs don't care. companies are spending on ai, but most of it comes from the cheaper models. give it a read ↓ pran.sh/blog/reliab...

pran.shCoding agents need receipts, not fansTeams will not trust coding agents because of launch hype. They will trust them when they can see what the agent handles well, where it fails, and how other engineers use it.

18 Aug

I've got a follow up from yesterday! 💡 I wrote about what sets the bar, and realized that if I have shared the problem, I also have to write about how can these projects show receipts. TL;DR | Nobody trust agents because of launch hype. New post → pran.sh/blog/trust-...

17 Aug

i’ve been noticing a pattern on Launch HN lately: it’s slowly turning into a coding agent directory. every week there’s another agent that opens PRs, fixes issues, plugs into Linear, runs in the background, or “ships like an engineer.”

17 Aug

a lot of them are genuinely impressive. But I don’t think the interesting question is “which one writes code fastest?” that part is getting cheap. the more interesting question is: which ones can earn enough trust that teams actually let them touch real work?

pran.shThe Next Coding Agent Moat Is TrustEvery week there’s another coding agent on Launch HN. The question now isn’t whether they can write code, but whether teams will trust them.

17 Aug

because the default competitor is not just another startup -- it’s, “I’ll just ask Claude in my editor.” wrote some thoughts on what I think actually matters for coding agents now: trust, distribution, workflow, narrow wedges, and taste. pran.sh/blog/hn-agents

14 Aug

Conversations from this week 👂 1/ Small-team pain in conversations -- AI speeds up frontend shipping, but creates more tiny UI bugs and designer/dev handoff friction. We're being sold, “AI builds it,” but the opp still lies in tools that keep taste, QA and iteration synced.

14 Aug

2/ AI API pricing is becoming product UX, builders care about cost predictability as much as model quality now. DeepSeek price hike + Gemini moving devs to prepaid both point the same way.

14 Aug

3/ Another YC coding agent(bullet) is positioning on speed, not “AI magic.” That’s the devtools market now - faster loops, fewer handoffs, clearer workflow ROI. If your agent pitch can’t explain the workflow win in one sentence, devs will bounce.

11 Aug

4th response saying the same. great time to be a community profesh in US, for sure. does anyone else see it too?

31 July

people are same everywhere 🫶

31 July

issa @figma socks day 🧦

30 July

watching odyssey while cooking, as nolan intended

30 July

dev meetups in 2026: hundreds watching a person asking claude to code

20 July

delighted to announce that my monthly ai usage limit on zoom has been reset after using it zero times last month 🎉

14 July

anyone using hermes as their daily driver with local models or claude out there? would love to know more about your setups

13 July

not me vibe-coding a grid only to add the snake game for nostalgia purposes on my website high score: 23 pran.sh/

12 July

two conferences, two weeks apart. very different audiences. At GITEX Europe, the conversations were about AI infrastructure, enterprise adoption, digital transformation, government initiatives, and where the industry is heading over the next decade.

One audience is thinking about strategy, investment, and transformation. The other is thinking about workflows, APIs, performance, and user experience. Attending both was a good reminder that the tech ecosystem is much bigger than the developer bubble most of us live in.

12 July

The most valuable insights usually happen somewhere in the middle. Certainly not around an F1 car, it was cool though.

You'd walk from a cloud provider booth to a startup founder pitching the future of an entire sector. At WeAreDevs, it was all about building. Editors, frameworks, open-source, developer tools, engineering culture, and the practical realities of shipping software everyday.

this is fun

new day, new adventure moderating stage 1 at WeAreDevs

27 Apr

Happening tomorrow at Mozilla Berlin! 🚨 🎤 Talk on Firefox Mobile and open source contributions 🗺️ Firefox Roadmap & upcoming features 🤝 Networking 🎟️ Raffle 🗓️ Apr 28 // 18:00 // Schlesisches Tor RSVP now! 👇 www.meetup.com/berlin-mozi...

25 Apr

MDN's old frontend had: ejected Create React App, a Webpack config nobody wanted to touch, CSS so entangled fixing one component broke others, and dangerouslySetInnerHTML as a core pattern. So the team rebuilt it from scratch. Deep dive worth reading 👀 developer.mozilla.org/en-US/blog/...

24 Apr

Something exciting is happening in Berlin on Tuesday 👀 We're hosting a community meetup at the Mozilla Office - talks, Q&A, networking, and a raffle with the Firefox iOS Team. 💟 🗓️ April 28 // 18:00-21:00 RSVP 👇 www.meetup.com/berlin-mozi...

We just published a deep dive into how we rebuilt @MozDevNet's entire frontend. And I'm so proud of what the team shipped. ♥️ developer.mozilla.org/en-US/blog/...

Leo McArdle wrote a fantastic walkthrough of the whole architecture. If you're curious about web components in production, modern build pipelines, or just want to see how a site serving 20M+ visitors rebuilt its frontend, this is worth your time.

Read it here 👇 developer.mozilla.org/en-US/blog/...

We just published a deep dive into how we rebuilt @MozDevNet's entire frontend. And I'm so proud of what the team shipped. ♥️ developer.mozilla.org/en-US/blog/...

The result? Dev server starts in 2 seconds. Each page only loads the CSS and JS it actually needs. Web components lazy-load in parallel. And our content authors can now write interactive examples directly in Markdown.

But here's what I love most: we used the web platform to build the site that documents it. Custom elements, Shadow DOM, progressive enhancement. The stuff we write about every day on MDN, we're now building with.

We just published a deep dive into how we rebuilt @MozDevNet's entire frontend. And I'm so proud of what the team shipped. ♥️ developer.mozilla.org/en-US/blog/...

Here's the short version: our old frontend was a React SPA that had accumulated years of tech debt. Webpack configs that nobody wanted to touch. CSS that was too entangled to scope properly. A dev server that took 2 minutes to start. So we rewrote it. From scratch.

Here's what we reached for 👇 → Lit and web components for all interactivity → Custom server components for static templating → Rspack instead of Webpack → Declarative Shadow DOM for instant rendering → Baseline to decide which web APIs we can ship with confidence

12 Mar

Stumbled on git-pulse's snapshot rankings and had to look up mdn/content. ✨ 4th in new committers per month (49) ✨ 5th in new committer percentage (48%) ✨ 5,009 total contributors For a documentation repo, that's wild. And we are actively growing. git-pulse.github.io/snapshots/

David Cramer wrote something that clarified a lot of the noise for me around MCP, skills, and agents. The key insight: these aren't competing ideas. They each solve different problems.

A simple way to put it is, "If skills teach you to cook, MCP provides the instruments that let you do it." What I appreciate most is the pushback on the instinct to declare one approach the winner. Every new thing doesn't have to kill the last thing.

cra.mrMCP, Skills, and AgentsIf you're not caught up on the current hypewave, coding agents like Claude Code and Cursor have latched onto skills recently. Like with all of these waves, everyone has immediately proclaimed they solve all problems, and all previous iterations of technology are no longer relevant. That's obviously not true, and I want to talk about my thinking behind them.

Sometimes they just... work together. 🤷‍♂️ I've been reading a lot more lately. And when something stays with me, I want to pass it along - for the community, for friends. Hope this one's useful for you too. cra.mr/mcp-skills-...

David Cramer wrote something that clarified a lot of the noise for me around MCP, skills, and agents. The key insight: these aren't competing ideas. They each solve different problems.

→ Skills are reusable prompt bundles. Lightweight, loaded on demand. They teach the agent how to do something. → MCP exposes tools - the instruments that let you actually do it. → Sub-agents are isolated workers with their own context. Think map/reduce for AI workflows.

10 Mar

As AI does more of the work, what matters most: Asking better questions. Not just what it returned. But whether you asked the right thing. That skill gets rusty if you stop practicing it. - Addy Osmani on critical thinking in the AI age