Andy McLoughlin @Bandrew

When we invested in Mirai Labs (@trymirai) earlier this year, they had a long-ish term plan to offer full-stack on-device inference infrastructure and the underlying models. I thought it would take them a year... In less than four months, their co-designed quant+inference stack for Apple Silicon shows 40–60% more tokens/sec vs llama.cpp and MLX at the same quality level. This week they published technical posts on quantization (trymirai.com/blog/quantizat…) and sparse buffers (trymirai.com/blog/sparse-bu…) which I encourage app builders and inference consumers to read. Awesome stuff from @Darkolorin, @dmitrshvets, and the Mirai team.

There's a reasonably sophisticated scammer hitting up my friends, colleagues, co-investors, and LPs from a random 332 number. It's not me. Which is obvious if you get into it with them as that start talking about crypto and blockchain and seminars and other such nonsense. Stay vigilent!

@thistrippjones Seed is dead, long live seed? 🤷🏻‍♂️

BREAKING: Apple is no longer the final boss every app you vibe code on Anything now ships to the Play Store too

@josephruscio @PeterJ_Walker Checks Notes Twitter remains undefeated

MIA is MIA

Since launching in 2015, @LaunchDarkly has been synonymous with enterprise-grade feature management (50 trillion feature evals per day, serving many of the most exciting companies in the world). Today they're announcing AgentControl, the first product as they become the runtime control platform teams trust to move quickly with AI. While staying in control of code (and agents) in production, of course. Very excited for LD's next chapter! 🚀

Edith Harbaugh @edith_h

3 weeks ago

I started LaunchDarkly because teams deserved better control over what they shipped. With AI, that matters more than ever. Today we’re launching AgentControl: real-time visibility and control for AI agents in production. ld.team/6W88xI

2 3 18 9K 8

Mirai Tech is now Mirai Labs (@trymirai) A full-stack frontier lab for on-device AI. Models, inference engine, quantization, and application layer, all co-designed from the hardware constraint up. Interactive AI is a continuous loop: parsing, validating, tool calls, rendering. That only works when latency disappears. Cloud models can't get you there as they're built for large batches, abundant memory, throughput as the metric. On-device inverts every assumption, requiring the architecture that was built from the hardware constraint up. The silicon is shipped. The software stack to match the chip doesn't exist yet. That's ours to build.

At Nuon, we eat our own dogfood. We run Nuon as a BYOC service in our customer's accounts. It's not easy, but it is insightful and makes us better. Here are some of the hardest challenges we see, which you might see as well when offering your BYOC:

Obligatory "I spoke on the big stage at Web Summit" photos

keyboard_arrow_left Previous keyboard_arrow_right Next

Had a lot of fun (really, actually) chatting with @kentlind on the Something Ventured podcast. We cover a lot of ground: the early days of seed investing (featuring folks like @jeff, @m2jr, @joshk et al), the state of seed today, Silicon Valley's "British invasion" (Kent's words, not mine), the great charitable work being done by Founders Pledge, San Francisco's resurgence (powered in no small part by @GrowSF), and - of course - AI. Lots about AI. somethingventured.us/uncork-capital…

Everything Also In French is one of my favorite things about visiting Canada

@TurnerNovak Is it the guy from Flybridge who became “the person on twitter” today

Every AI founder has the Claude Spookies right now. Will Claude’s next release kill our product? I studied the fastest-growing vertical AI companies (e.g. Cursor, Sierra, Legora) to uncover the 3 moats that actually hold. In February 2026, one Anthropic update made Ryze AI, a startup with a 70% deal close rate, irrelevant overnight. The founder posted "Claude killed my startup." Every founder felt it. The week Anthropic released Cowork, $1 trillion of value was wiped from the stock market. Last week, Anthropic released 10 finance agent templates and FactSet dropped 8.1% the same day With OpenAI re-entering enterprise, model providers will keep moving up the stack as competition heats up. At the same time, vertical AI companies are growing at insane rates. @cursor_ai added $1B in ARR in a single quarter (Coding). @WeAreLegora hit $100M ARR in 18 months and @evenuplaw 4X'ed revenue to $100M+ (Legal). Abridge went from $8M to $117M ARR in 15 months (Healthcare). @SierraPlatform hit $150M in 2 years (Customer service). These companies share 3 moats that make customers choose them over Claude: 1. Proprietary data loops that compound Abridge is built on 50M+ clinical conversations. EvenUp built a proprietary settlement database of 200K+ cases and millions of medical records. Clay has a data marketplace that's needed despite LLMs commoditizing outbound. Sierra's Journeys codify each customer's voice, procedures, & brand that compound over time. These data investments are too fringe for LLMs, but they matter to busy customers who need time to value. 2. UX is built for the customer’s workflow Abridge saves physicians 2 hours a day, because it lives inside Epic's native interface. Legora lives inside Word and Outlook, where lawyers already work. Its tabular view makes bulk actions like reviewing 500 customer contracts during an M&A deal easy. Cursor is so good at embedding into the org's workflows and context, making it easy to go from feedback to PRs. The workflow-native UX, paired with FDEs who know the buyer's domain, are why vertical AI keeps winning deals over LLMs. 3. Platform leverage that compounds The best vertical AI companies are consolidating workflows and becoming the platform other workflows route through. Legora is now an agentic OS for legal work, embedding inside Microsoft Word and law firm document management systems. Abridge started with clinical notes but is expanding into orders, prior auth, and revenue cycle, becoming infrastructure across 250+ health systems. While general-purpose models will accrue more integrations, they can't replicate a platform that third parties have already built their businesses on. At @mutinycorp we're investing in all three: workflow-native UX, proprietary contact-level analytics from owning the asset layer, and a platform play we'll share more on soon. The Claude Spookies are real, but so is the demand. The $10B+ vertical AI winners will nail time-to-value and get harder to rip out the longer you use them.

“I have a ton of friends who also work in venture”

PsyPost.org @PsyPost

a month ago

A new study shows that individuals with similar levels of autistic traits are naturally drawn to one another. Brain scans reveal they use alternative, highly effective neural strategies to connect, challenging traditional deficit-based models of autism. dlvr.it/TSSN7t

703 2K 36K 23.0M 16K

(No doubt this tweet will get some heat from the crazies who want to keep things dysfunctional)

San Francisco is on the right track and we can be proud of how the city is bouncing back. But there’s still a lot to do: downtown recovery, more housing, clean streets, public safety. The @GrowSF guide has pretty much nailed who you should vote for in order to keep things going.

Perplexity, Netlify:

Ara Kharazian @arakharazian

a month ago

Ramp TOP SAAS VENDORS MAY 2026 @tryramp 1. FIGMA is a fastest growing vendor, and the only publicly traded company on our list. Now competing with Claude Design. 2. AI inference platforms grew as businesses opted for cheaper models to combat exploding token budgets. 3. It's a

23 22 209 217K 157

Excited to join @meliarobin, @Bandrew, @ilyaf and Colin Tobias on stage at TechGC Nexus tomorrow discussing the future of work and legal AI! Cannot promise that Jude Law vs. Harvey Specter won't come up...

What a great first five years, and what a fun way to celebrate.

Wander @wander

a month ago

Today, Wander turns 5 years old.🎉 To celebrate you helping us make it here, we’re giving away a free trip to one of our 5 original properties. To enter, simply reshare this post and tag the person you’d want to go on an adventure with in the replies.

34 57 151 1.4M 16