Hello World

99% of Ramp uses ai daily. but we noticed most people were stuck — not because the models weren't good enough, but because the setup was too painful and unintuitive for most. terminal configs, mcp servers, everyone figuring it out alone. so we built Glass. every employee gets a fully configured ai workspace on day one — integrations connected via sso, a marketplace of 350+ reusable skills built by colleagues, persistent memory, scheduled automations. when one person on a team figures out a better workflow, everyone on that team gets it and gets more productive. the companies that make every employee effective with ai will compound advantages their competitors can't match. most are waiting for vendors to solve this. we decided to own it.

It has been such an honor to work with so many amazingly talented people

Software horror: litellm PyPI supply chain attack. Simple `pip install litellm` was enough to exfiltrate SSH keys, AWS/GCP/Azure creds, Kubernetes configs, git credentials, env vars (all your API keys), shell history, crypto wallets, SSL private keys, CI/CD secrets, database passwords. LiteLLM itself has 97 million downloads per month which is already terrible, but much worse, the contagion spreads to any project that depends on litellm. For example, if you did `pip install dspy` (which depended on litellm>=1.64.0), you'd also be pwnd. Same for any other large project that depended on litellm. Afaict the poisoned version was up for only less than ~1 hour. The attack had a bug which led to its discovery - Callum McMahon was using an MCP plugin inside Cursor that pulled in litellm as a transitive dependency. When litellm 1.82.8 installed, their machine ran out of RAM and crashed. So if the attacker didn't vibe code this attack it could have been undetected for many days or weeks. Supply chain attacks like this are basically the scariest thing imaginable in modern software. Every time you install any depedency you could be pulling in a poisoned package anywhere deep inside its entire depedency tree. This is especially risky with large projects that might have lots and lots of dependencies. The credentials that do get stolen in each attack can then be used to take over more accounts and compromise more packages. Classical software engineering would have you believe that dependencies are good (we're building pyramids from bricks), but imo this has to be re-evaluated, and it's why I've been so growingly averse to them, preferring to use LLMs to "yoink" functionality when it's simple enough and possible.

Replit now deploys directly to Databricks. Your apps run inside your Databricks environment while inheriting its security, governance, and data access. Beta is live. Enterprises are already building with it and seeing massive acceleration in BI and internal tools.

@thsottiaux To think that I remember when Codex looked like this: openai.com/index/openai-c… And now it's reached it's product market fit. Nearly 5 years into it. It's come a long way and I believe only getting started.

At the current codex growth pace, we will owe you all another reset in less than two weeks. Would be funny to reach a state where these happen more than once a week before we reach 10M. Scrambling to get the capacity online.

If you want to understand the future of AI, watch this discussion with OpenAI’s chief scientist

"We have never seen anything like this." @Jason Anthropic's revenue ramp is through the roof. Why? @CarinaLHong says it is because of their coding models. OpenAI thinks of coding as just another vertical, and this may have left them in the dust.

My latest: OpenAI Highlights Massive Compute Advantage Over Anthropic to Investors taekim.substack.com/p/exclusive-op…

Judging by my tl there is a growing gap in understanding of AI capability. The first issue I think is around recency and tier of use. I think a lot of people tried the free tier of ChatGPT somewhere last year and allowed it to inform their views on AI a little too much. This is a group of reactions laughing at various quirks of the models, hallucinations, etc. Yes I also saw the viral videos of OpenAI's Advanced Voice mode fumbling simple queries like "should I drive or walk to the carwash". The thing is that these free and old/deprecated models don't reflect the capability in the latest round of state of the art agentic models of this year, especially OpenAI Codex and Claude Code. But that brings me to the second issue. Even if people paid $200/month to use the state of the art models, a lot of the capabilities are relatively "peaky" in highly technical areas. Typical queries around search, writing, advice, etc. are *not* the domain that has made the most noticeable and dramatic strides in capability. Partly, this is due to the technical details of reinforcement learning and its use of verifiable rewards. But partly, it's also because these use cases are not sufficiently prioritized by the companies in their hillclimbing because they don't lead to as much $$$ value. The goldmines are elsewhere, and the focus comes along. So that brings me to the second group of people, who *both* 1) pay for and use the state of the art frontier agentic models (OpenAI Codex / Claude Code) and 2) do so professionally in technical domains like programming, math and research. This group of people is subject to the highest amount of "AI Psychosis" because the recent improvements in these domains as of this year have been nothing short of staggering. When you hand a computer terminal to one of these models, you can now watch them melt programming problems that you'd normally expect to take days/weeks of work. It's this second group of people that assigns a much greater gravity to the capabilities, their slope, and various cyber-related repercussions. TLDR the people in these two groups are speaking past each other. It really is simultaneously the case that OpenAI's free and I think slightly orphaned (?) "Advanced Voice Mode" will fumble the dumbest questions in your Instagram's reels and *at the same time*, OpenAI's highest-tier and paid Codex model will go off for 1 hour to coherently restructure an entire code base, or find and exploit vulnerabilities in computer systems. This part really works and has made dramatic strides because 2 properties: 1) these domains offer explicit reward functions that are verifiable meaning they are easily amenable to reinforcement learning training (e.g. unit tests passed yes or no, in contrast to writing, which is much harder to explicitly judge), but also 2) they are a lot more valuable in b2b settings, meaning that the biggest fraction of the team is focused on improving them. So here we are.

🚨 𝟱𝗛𝗥𝗦 & 𝟮 𝗦𝗣𝗢𝗧𝗦. 𝗬𝗢𝗨𝗥 𝗖𝗟𝗜𝗘𝗡𝗧𝗦 𝗔𝗥𝗘 𝗣𝗔𝗬𝗜𝗡𝗚 𝗙𝗢𝗥 𝗖𝗟𝗔𝗨𝗗𝗘 𝗖𝗢𝗗𝗘. 𝗬𝗢𝗨 𝗖𝗔𝗡 𝗗𝗘𝗣𝗟𝗢𝗬 𝗔 𝗕𝗘𝗧𝗧𝗘𝗥 𝗠𝗢𝗗𝗘𝗟 𝗙𝗢𝗥 𝗙𝗥𝗘𝗘. 𝗣𝗥𝗜𝗖𝗘𝗦 𝗚𝗢 𝗨𝗣 𝗜𝗡 𝟱 𝗛𝗢𝗨𝗥𝗦. 𝗦𝗧𝗜𝗟𝗟 𝗢𝗡 𝗧𝗛𝗘 𝗢𝗨𝗧𝗦𝗜𝗗𝗘? 🚨 → skool.com/ai-profit-lab-… 🚨 𝟯 𝗖𝗟𝗔𝗜𝗠𝗘𝗗. 𝟮 𝗟𝗘𝗙𝗧. 𝟴-𝗛𝗢𝗨𝗥 𝗔𝗨𝗧𝗢𝗡𝗢𝗠𝗢𝗨𝗦 𝗖𝗢𝗗𝗜𝗡𝗚 𝗔𝗚𝗘𝗡𝗧. $𝟬 𝗧𝗢 𝗥𝗨𝗡. → skool.com/ai-profit-lab-… 🚨 𝟮 𝗦𝗣𝗢𝗧𝗦. 𝟱 𝗛𝗢𝗨𝗥𝗦. 𝗧𝗛𝗘 𝗣𝗟𝗔𝗬𝗕𝗢𝗢𝗞. → skool.com/ai-profit-lab-… Let me show you what "8 hours autonomous" actually means in dollars. A client pays you $10K to build them a custom GLM 5.1 + OpenClaw coding agent. You deploy it in 2 hours using the delivery checklist. Their agent now runs overnight. Plans. Codes. Tests. Fixes. Optimizes. By morning, their dev team wakes up to work that would've taken a week. No subscription bill. No token costs. No per-seat pricing. Just the one-time build fee they paid you. Meanwhile, their competitor is still paying $200/month per developer for Claude Code. $20/month per seat for Cursor. Thousands for dev tooling. Your client is 6 months ahead of schedule and saving five figures a year. Then they refer you to 3 other CTOs. One client. That's how the math works. 👉 Get the playbook before the last 2 spots fill → skool.com/ai-profit-lab-… 🎯 𝗧𝗛𝗘 𝗣𝗟𝗔𝗬𝗕𝗢𝗢𝗞 (𝟮 𝗦𝗣𝗢𝗧𝗦): ✅ Free Stack Setup (GLM 5.1 + OpenClaw, 8-hour autonomous) ✅ 5 Services to Sell (coding agents, audits, refactors, bug-fix, dev tooling) ✅ Pricing Sheet ($1.5K-$15K tiers) ✅ Sales Scripts ("8 hours of autonomous coding for free" angle) ✅ Client-Finding Playbook (5 free channels) ✅ Delivery Checklist (2-hour setups) 👉 Get the playbook and land your first client this week → skool.com/ai-profit-lab-… 📦 𝗙𝗨𝗟𝗟 𝗦𝗧𝗔𝗖𝗞: ✅ OpenClaw + 6-hour course + 100+ workflows ✅ Claude Blueprint + Agent Teams + Auto-Mode ✅ Hermes roadmap + Quick Deploy ✅ AI Avatar Clone System + NemoClaw Security + Manus Workflows ✅ $5K High Ticket Formula + funnels ✅ 105 Agency-Level Prompts 👉 1,000+ n8n automations ✅ 5 live calls/week ✅ Daily Q&A ✅ Unlimited support ✅ 30+ new trainings/week ✅ $8K+ monthly giveaways ✅ 2,700+ members 𝗣𝗥𝗜𝗖𝗘𝗦 𝗚𝗢 𝗨𝗣 𝗜𝗡 𝟱 𝗛𝗢𝗨𝗥𝗦. Playbook gone at midnight. 2 spots. 👉 Close the gap → skool.com/ai-profit-lab-…

I’ve locked in 50+ B2B influencers over the next 3 weeks to promote Gojiberry.ai. So far, the results are looking very strong. Next step: I’ll use our own software to run warm outreach, reaching out with ultra-personalized messages based on what they do and real buying signals. If you want the full list of 200+ B2B influencers I’ve curated, share this post and comment “LIST”. I’ll send it straight to your inbox.

I know it's self serving to say, but man I would've killed for a resource like Tinker and the tutorials, the cookbook, etc back when I was in undergrad. Following @karpathy blogs and training RNNs on a crappy Acer *was* fun, but doing bigger things with less setup is such a boon

As MCP adoption grows, teams are moving quickly to secure how LLMs connect to tools and data. We put together 7 best practices to help you: - Lock down supply chains - Enforce least privilege - Add human oversight

Computer now connects with Plaid to link bank accounts, credit cards, and loans. Track spending in detail, build custom budget tools, and visualize your net worth alongside your investment portfolio.

A Bloomberg Terminal costs $30,000/yr and still can't do a fraction of what Perplexity Computer just launched today 💻 It now connects directly to your bank accounts, credit cards, loans, and brokerage accounts through Plaid. Your full financial picture, from monthly spending to net worth to individual stock positions, sitting on top of 40+ live finance data sources including SEC filings, FactSet, S&P Global, and Coinbase. Every dollar you earn, spend, owe, and invest, cross-referenced against institutional-grade data in real time. You can walk up to this thing and say: → Run a risk analysis on my portfolio against the current tariff environment → Show me where my spending spiked last month and what's driving it → Build a net worth dashboard that tracks everything in one place → Flag any holdings that overlap with what insiders have been selling this quarter And it just does it. Pulls from your linked accounts, cross-references SEC filings, builds the output, and delivers a finished product. The system running underneath is Perplexity Computer. It orchestrates 19 models simultaneously, breaks any goal into subtasks, spins up specialized agents for each one, and keeps working after you walk away. One model handles the reasoning. Another does the research. Another writes the code. Another builds the visualization. All coordinated automatically. Last month they launched with brokerage data only and someone built a Bloomberg Terminal clone in a single afternoon. That post did 7.5 million views. Now they've expanded to your entire financial life: checking, savings, credit cards, loans, and investments all in one place. Wall Street pays $30K a year for a terminal with 30,000 function commands built over four decades. It won't replace Bloomberg for institutional traders executing billion-dollar orders. But for everyone else, the gap just got a lot smaller.

Brian Armstrong explains how he built Coinbase on nights and weekends while working at Airbnb Brian first advises those who are currently employed to not build your project on company hours or on your company laptop: “If you build it on company time or on the company hardware, the company probably owns the IP.” Then he describes his schedule for working on Coinbase while still working full-time at Airbnb. “I would often work [at Airbnb] until 7pm. I’d come home, eat dinner, and then I would work from 8pm to midnight. I would do that maybe 3-4 days a week on weekdays. And then on the weekend I’d work Sunday afternoon for 7-8 hours.” Brian did this consistently for about a year and a half until Coinbase was far enough along for him to get seed funding from Y Combinator. “It sucked. I mean I was tired after the full day of work [at Airbnb]. But this is where determination comes in… At that moment in time, I was in my late 20s, and I was like, ‘I really want to try to build something important in the world.’” When asked how he maintained friendships during this time, Brian replies: “I was pretty intense about it. I would say I sacrificed friendships for it. It’s not like I was just never responding to people, but I’ve seen this happen to various people. They get to a certain point in their life. Sometimes they turn a certain age where they thought they would have more done by then or maybe someone in their family passes away and they’re like, Oh my god, time is finite. It’s precious. And something happens where they’re like, ‘I’m going to get this done, no matter the cost.’” Brian tells those out there who might be in a similar situation: “Go hard at it. Finish your book. Launch your thing. Just start doing stuff - and even if you don’t know what to do, just do anything, because action will produce information and it’ll help you get to the right thing.” Video source: @StevenBartlett (2022)

Jack Dorsey’s #1 rule for pitching your startup: “Show them something that works” Jack Dorsey, Jim McKelvey, and another engineer built the initial prototype for Square in a month. As Jack recalls: “I could actually swipe a card, generate an electronic receipt via email, and then send it out to a person. I loved this because I would go around to all these angel investors and VCs and charge them $5-50 to show them my new idea.” This prototype was critical in helping the company raise $10 million in initial funding from Khosla Ventures. And Jack believes it’s a lesson all founders should take to heart: “The thing that really inspires people is a working product. When you’re pitching someone, the best thing you can do is show them something that works.” Jack continues: “We did this with Twitter. We had a great number of users… and we had investors who were coming to us who were already users of the product. Their families were users of the product. So the story became very easy to tell and they could easily see why this was something that was powerful.” Video source: @Stanford (2011)

My besties at @greatchatpod had me on yesterday to talk about new media open.spotify.com/episode/5IjgfP… Key takeaway: AI is finally paying for content!