Built https://xitter.cf/t.co/HPoFyD35a0 from zero to acquisition. Sharing everything I learned about what breaks (and what works) at early-stage startups.

Joined October 2014
Replying to @PaulYacoubian
@PaulYacoubian and I are excited to share what we've been working on over the last few weekends!
Launching Copy.ai If you hate writing marketing copy, it will blow your mind. We improved on GPT-3 to write copy for any product and for any audience in seconds. Try it free for 7 days. Unlimited usage. No beta invitation needed.
18
9
4
195
Nikos is one of the most thoughtful high slope entrepreneurs I’ve met and it shows in his products! Check it out!
I’ve been building with LLMs since my first startup in 2022, and for years I waited on the sidelines for someone to solve memory. Nobody did, so I spent the last 4 months building it: a shared learning layer that gives every agent on your team the context your best engineer carries around. Today we’re opening it to everyone: Glen • 21% faster to a merged change while using 1/3 fewer tokens. • One agent learns something, every agent on your team knows it (harness agnostic, cross teammate) First 15,000 memories are free, we are building the shortest path to a truly compounding org.
1
2
501
This looks insane!
🚀 We’re thrilled to announce the release of Lucky Engine 2026.3! "The Matrix for Robots" is now 100% Agentic 🤖✨ To train autonomy, you need a "Matrix" a virtual playground where they can learn before touching the physical world. Today, we are dropping Lucky Engine 2026.3, and we think that it redefines how synthetic environments are built. • Real-time ray tracing & path tracing • 100% agentic (AI agents fully control the engine) • Upload a photo → instant data generation for your policies • Graph-based procedural generation, AI generated determinism • No Blender required. Here’s what’s inside 🧵 👇
1
6
757
GLP-1 for your AI bill. On the streets of San Francisco today. glp1.orchestra.ai
2
4
1
24
845
Chris Lu retweeted
Your AI is vaporware without deep integrations with your customers’ systems of record. Introducing @WithAmpersand: integration infrastructure for enterprise agents. We power 11x, Orb, and Square's ability to take agentic action in systems of record like Salesforce, SAP, NetSuite, and Workday. Ask anyone serious about building AI and they’ll tell you: - The SaaSpocalypse didn’t happen. The world depends on CRMs, ERPs, HRISs, and ITSMs. - Your customers customized their deployments beyond recognition. - Systems of record companies are basically monopolies. They never had to make their APIs and MCPs user-friendly. - Docs don't explain half the weird edge cases you'll hit. - One bad write can blow up your pilot or renewal. - Once you finally get it working, someone changes a field and it breaks again. The world's data lives in structured databases. AI needs a translation layer to work with it. So, for your AI to truly transform the way enterprises work, you'll need deep integrations built for: - scoped permissions - bi-directional actions - real-time speed for agents - custom objects, fields, and workflows for each of your customers We built Ampersand to power the future of software. AI didn't trivialize writing integrations. It made them the critical path. I believe that deeply, so I didn't make a launch video about Ampersand. Instead, it's the best builders I know explaining just how big of a challenge this is.
168
130
146
551
848,248
I've looked at so many iPaaS, ETL, etc. platforms. Ampersand is the absolute best for deep product integrations. They allow you to create your own unified API across the hardest systems (CRM, ERP, etc.) Congrats on the Series A!!
Your AI is vaporware without deep integrations with your customers’ systems of record. Introducing @WithAmpersand: integration infrastructure for enterprise agents. We power 11x, Orb, and Square's ability to take agentic action in systems of record like Salesforce, SAP, NetSuite, and Workday. Ask anyone serious about building AI and they’ll tell you: - The SaaSpocalypse didn’t happen. The world depends on CRMs, ERPs, HRISs, and ITSMs. - Your customers customized their deployments beyond recognition. - Systems of record companies are basically monopolies. They never had to make their APIs and MCPs user-friendly. - Docs don't explain half the weird edge cases you'll hit. - One bad write can blow up your pilot or renewal. - Once you finally get it working, someone changes a field and it breaks again. The world's data lives in structured databases. AI needs a translation layer to work with it. So, for your AI to truly transform the way enterprises work, you'll need deep integrations built for: - scoped permissions - bi-directional actions - real-time speed for agents - custom objects, fields, and workflows for each of your customers We built Ampersand to power the future of software. AI didn't trivialize writing integrations. It made them the critical path. I believe that deeply, so I didn't make a launch video about Ampersand. Instead, it's the best builders I know explaining just how big of a challenge this is.
2
4
27
2,148
Listened to two podcasts, one with @dougleone and one with Mike Moritz They both said that their children / grandchildren gave them the most happiness and meaning
Lost in the narrative of Silicon Valley is what really matters. Yes money is a thing. And a billion dollars is a lot. But over the course of a lifetime what really matters are family, friends, relationships.
1
5
452
Chris Lu retweeted
Excited to announce Volantis's $88M Series A. We are solving Al's memory bottleneck by using optics, enabling chips with huge amounts of fast & cheap memory. By boosting both the memory bandwidth and capacity per chip by orders of magnitude, we enable ultra-fast inference (up to 10,000 tps/user) for large models (>10T) - with low $/tok to boot. Initially, this will enable insanely fast agents - think coding agents that finish in minutes or even seconds instead of hours. More excitingly, optics is a fundamentally scalable way to increase memory systems. Not 2X/year, but by orders of magnitude across new generations. This will enable a structurally new Al industry, including restarting scaling laws, holding entire repos in context windows & more. Our team has pioneered many core semiconductor technologies: the 1st CoWoS product, early HBM, the 1st silicon photonics CPO systems, the 1st high volume tunable VCSELs, the 1st processors to directly communicate using light & more. We’ve already sent data >10× farther than equally tiny electrical wires inside a chip package. Our next iteration is already taped out and targets world-record bandwidth density over relevant distances, read more: volantissemi.ai/news-insight…
212
182
124
1,349
1,014,499
Chris Lu retweeted
This is Kled V3. We've solved data collection for artificial general intelligence. Any consumer dataset that can exist, can now be collected from physical reality in under 72 hours. All powered by the largest and most comprehensive data application layer on the planet. (Thread)
184
103
67
2,968
3,812,288
Am I the only one seeing this?
6
447
Jev by @typesafeai went from 0 to $100m in 7 DAYS...WTFFFF
The @BostonCollege Investment Committee (an LP and my beloved alma mater) asked for a few thoughts on what's happening in AI. I recorded a test run yesterday morning and then shared it with my partners, who encouraged me to share it more broadly... so here you go! This is not a sales pitch, it's just a reflection on what we're seeing. And it wasn't intended to be shared, so please pardon the rough edges. loom.com/share/c016702964a04…
10
7
4
312
81,584
Chris Lu retweeted
Can Jev replace ATC aka air traffic controllers? It did for me. You can try to! Trigger a Mayday, close a runway, invite storms, its instant on how it adapts! Parallel runways too. PS: No plane was hurt while building this. Try it out: atc.vibe.agnost.ai/
9
2
3
38
3,956
YC's newest batch has more startups building RL environments than startups selling AI to sales teams. I went through all 236 companies in S26. Here's what I found:
4
3
1
60
16,733
Right when GPT-3 launched, people who emailed Greg Brockman were able to get off the waitlist if they had an interesting idea! Gave us a 1 year head start for CopyAI!
warm intros are low agency. you can just reach out
5
1
41
4,476
Chris Lu retweeted
Stop using your agent logs just for debugging. Use them to train your own model. Today we're launching Agnost AI (YC S26)'s first model: agnost-*******-0.1 Trained for our first customer on their existing production traces & it beat the frontier: +22.9% task success −90.2% latency −94.5% cost If you ever thought of fine-tuning your own model, we'll do it for you, talk to us!
113
69
53
446
697,021
There is a lesson somewhere here: surround yourself with people who deeply believe in you can increase probability of success
1
6
402
Excited to announce that @aamirpoonawalla and I are launching @understudylabs as part of @ycombinator 's S26 batch. We help you cut your company’s Anthropic bill by 80% with no loss of performance. Understudy is an inference cloud. We capture your AI traces, train smaller models, and automatically deploy them (only when they outperform the expensive model you are currently paying for). Instead of renting the same frontier model forever, your AI infrastructure gets cheaper and better every time it runs.
49
20
8
409
50,377
In March 2024, at the age of 35, I was diagnosed with stage IV rectal cancer with a metastasis in my liver. This was a shock: I had no family history, and none of the doctors suspected it. After two years of chemotherapy, radiation, and multiple surgeries, I'm still not cured. Now I'm asking for help to get a personalized cancer vaccine. This is my story so far. Following diagnosis, I was treated with total neoadjuvant therapy at UCLA. I had 6 weeks of radiation and chemotherapy (Xeloda). This was followed by 6 rounds of oxaliplatin infusions and Xeloda. I tolerated this unusually well. During this period, I managed to hike up to over 11,000ft multiple times in the Sierras, and I wrote the main text of a 72 author meta-analysis paper and submitted it to Nature where it is still in review. I showed a good clinical response, and at the end of October I had surgery: I was under for 9 hours, and they cut out a bit of my rectum, 1/3rd of my liver, and gave me a temporary ileostomy. Waking up from this was probably the strangest experience of my life: I felt more machine than man, and I had wild hallucinations form such a large dose of anesthesia. It took a while to feel OK after that surgery. However, I still had the ostomy. I had that reversed at the end of January in another surgery, which took a greater toll on me than I expected. I only just started to feel OK again and I got bad news: I got a positive circulating tumor DNA test (ctDNA), the signatera test from natera. Unfortunately, the ctDNA carried on rising, and in July 2025 a new lesion appeared in a lymph node near my liver (periportal). I was treated with irinotecan and panitumumab infusions, which showed a good response, followed by live MRI guided SBRT radiation treatment to the lesion. However, after finishing this treatment, the ctDNA remained positive and began rising again. I restarted irinotecan and panitumumab infusions, which brought ctDNA back to negative (signatera), although it remained positive on the more sensitive Personalis assay. During this period I had begun to develop stomach pain and digestive symptoms. Then I collapsed vomiting blood while out in Santa Monica and was rushed to the ER. They found out I had a duodenal ulcer that had bled. I had some blood transfused, and I was sent home to complete an intensive course of antibiotics for H. Pylori infection. However, after about a week, I started to get very weak. I collapsed again and went back to the ER in an ambulance. They found my hemoglobin was 5g/dL, a dangerously low level (normal is over 13.5). Multiple blood transfusions didn't bring it up and I collapsed again in the night after passing a large amount of blood. They took me to intensive care and performed an emergency endoscopy where they clipped and sprayed the ulcer. Thankfully this worked and I was discharged after another 3 nights in hospital. Unsurprisingly, nearly bleeding to death wasn't good for me, and my ctDNA began to spike rapidly. I had a PET-CT and it revealed 3 new disease sites in abdominal lymph nodes. I restarted irinotecan and panitumumab infusions after barely recovering from the ulcer incident. This showed an excellent response, with the sites resolving on imaging, although ctDNA remained positive. I've now started bot/bal immunotherapy, a not yet approved immunotherapy treatment that is the first of its kind to show strong results in my kind of cancer, micro-satellite stable colorectal cancer. This was possible thanks to a donation from an anonymous benefactor and supporter of free and open inquiry in science. Hopefully this treatment will be effective, but it is unlikely to be enough on its own to prevent future disease recurrence. I am asking for help to get a personalized neoantigen cancer vaccine, which could substantially increase my chances of surviving this deadly disease that is likely incurable under standard care. While cancer vaccines have shown promising results, they are not yet approved, meaning one has to pay for the design, manufacture, and administration of the vaccine. The total cost of which is $104,133, beyond my means or those of my family. Any help would be greatly appreciated! I've started a GoFundMe here: gofundme.com/f/cancer-vaccin…. Please donate and share with anyone who might be interested in helping. I've carried on working throughout this period and I'm still running a research group in statistical genetics at UCLA. I'm hoping to continue making contributions to science for as long as I can, hopefully longer with your help.
157
428
89
2,356
879,416