Most people think paying $200/month means their AI access is guaranteed OpenAI just reminded everyone otherwise → Reports said ChatGPT Pro limits were getting cut in half → Users pushed back → Tibo stepped in and doubled them instead A win for users but the bigger lesson? You're paying for access, not ownership Your AI limits can change overnight https://t.co/sY4gNd5QOS https://t.co/lDYRRQdFNY
Meta is moving faster than I expected After Muse Spark, seeing Meta push into competitive territory across coding, financial analysis, legal tasks and agentic work is interesting What stands out isn't just the benchmark numbers it's how quickly the competition is heating up And honestly for me that's a good thing No single lab should get too comfortable and with the constant push users are the ones who stand to benefit the most.
Most people use AI to finish tasks faster. Dan Martell is building a system where AI agents handle entire workflows, pass work between each other, and operate like a team. His setup combines 3 things: a computer-controlling assistant, specialized AI agents, and a shared AI memory. Here's how they work together. → ChatGPT Astra acts as his chief of staff, controlling his computer, accessing files and coordinating tasks. → GrokBot provides specialized AI agents for sales, finance, email and other workflows. → Apex stores his knowledge, decisions and preferences so his AI setup retains context even when he switches models. Here’s how they connect: Say Dan needs a financial report. Astra gathers the receipts from his computer, hands the information to a finance agent in GrokBot, then helps deliver the finished report through Slack. Each tool has a role. Astra coordinates, GrokBot’s agents do specialized work and Apex provides the shared knowledge. His philosophy? Own the brain. Rent the models. The real advantage isn't having AI tools. It's building a system where they can work together to get things done. Bookmark this.
Grok @bot the agent SpaceX uses won’t stick to one model. It picks whatever backend is best for the task: >Opus for reasoning >Midjourney for images >Suno for music >Other APIs when they fit better The bot is the chef while the models are the knives. If Opus is sharper today use it. If a Grok model is better tomorrow you switch. No single lab wins every task. Forcing one model to do all of them is how you get average work with a familiar name on it. Elon just showed what model agnostic AI actually looks like.
Order books shouldn’t have to wait for blocks to move And custody shouldn’t mean blindly trusting an operator The sweet spot is pretty simple: Match trades off-chain for speed, then settle them atomically on-chain. That’s the approach @perpdotcom is taking fast fills without giving up trustless settlement.
