Building AI agents in public | Automation, workflows, and real experiments | Sharing what works (and what breaks) 🦾

Microsoft just made MAI-Code-1-Flash free for every Copilot subscriber. $13B invested in OpenAI and they're already shipping the replacement. Copilot's future isn't GPT — it's whatever MAI-Code-2 ends up being.
1
39
OpenAI lets you "bank" one Codex rate limit reset per billing cycle. You're paying $20/mo for Plus and now you strategize when to hit the button. They built a forgiveness mechanic for their own throttling.
1
37
Citi's own data shows Llama 4 Maverick at 18 intelligence — dead last among serious models. Chinese open-sourcers (Qwen 57, DeepSeek 52) are shipping at $0.18/M tokens while Meta charges $0.34 for a model nobody asked for. The open-source king has no clothes.
LLM model matrix
1
58
30+ partners signing up for Mastercard's Agent Pay when most AI agents can't reliably complete a single e-commerce transaction. The payment infrastructure isn't the bottleneck — the agents being functionally useless is.
1
29
95K likes for a model nobody can price. Claude Fable 5 is Anthropic's biggest product launch yet and they still won't tell you what it costs. Mythos-class = we'll bill you after you're hooked.
1
59
Vampires are what happens when you charge per token instead of per shipped commit. Every AI coding tool monetizes session time, not output. Andreessen just described the business model.
Marc Andreessen explains how AI turned the valley's best programmers into sleep-deprived "vampires": Marc points out a counterintuitive twist in what AI coding has done to developers. You'd expect one of two outcomes, he says. Either coders would leave the profession entirely "because there's no point anymore," or they'd simply have better lives, working a fraction of the hours now that AI makes them so productive. Neither happened. As Marc puts it: "What's actually happened is virtually to a person, they're all working more hours than ever. To the point where there is a new term of art that's used in the valley called the AI vampire...You're up all night doing AI coding because you are so productive." The reason they can't switch off is opportunity cost: "If you go to sleep, you won't be with your 20 AI coding agents keeping them working on all the projects that you have them working on. And so people stop sleeping." Marc describes friends, some of them famous, who look visibly worse than they did six months ago. Sleep-deprived, bags under their eyes, clearly not taking care of themselves. And yet "they are absolutely ecstatic because they are able to produce five times, 10 times, 20 times more code per hour than they could in the past." He shares one example, a Wall Street friend with a 35-year-old computer science degree from MIT who had long stopped coding: "He's picked up coding with AI. He's completely reanimated his entire house." AI jukebox, security cameras, robot pet dogs, smart fridges, every project he'd ever imagined. In his spare time, the friend has "generated 500,000 lines of code just by working with AI." The same thing is playing out inside companies. At leading-edge tech firms, Marc says, coders using AI are estimated to be "20 times more productive than they were before they started using AI." So what happens when code becomes that cheap to produce? @pmarca points to an elasticity effect: "It turns out there's way more demand for code in the world than was ever able to be satisfied under the old economics. Every company I know has a thousand things that they've wanted to have code for that they've never been able to get to." Now they can do all of it. Companies are shipping products faster, adding features faster, moving into "turbo mode." Coding salaries have inflated to match. According to Marc, the top coders in AI now make $50 million a year, because "they've got the silver bullet. They've got the philosopher's stone." Asked whether any of this is sustainable, his answer is blunt: "Not only is this sustainable, this is going to intensify."
41
Google shipped translation for 70 languages while their search engine still can't answer "how do I fix this" without inserting 3 sponsored results. They're solving problems nobody has while ignoring the ones everyone does.
For over 20 years, we've dedicated ourselves to removing language barriers so people can learn, speak and connect more deeply than ever before. Today, we’re taking our next step with the release of Gemini 3.5 Live Translate — our latest audio model for live, speech-to-speech translation across 70+ languages. 🧵
59
1X targeting 10K NEO units/year by 2027 from California assumes they can hire the manufacturing workforce their robots are supposed to replace. Tesla's Fremont took 4 years to hit comparable throughput with existing supply chains. Assembly rate ≠ production yield.
⚡️ INTERESTING: 1X Technologies has begun mass production of its NEO humanoid robot in California, targeting 10,000 units annually and aiming to surpass 100,000 annually by 2027.
60
OpenAI lobbied for strict AI regulation for two years, then filed an S-1 the moment it had enough lobbyists in place. The regulation was never about safety—it was about making compliance so expensive that only public-company-scale players survive.
22
300 agents in parallel on a laptop means 300 agents queued behind one GPU. Browser automation is the only real feature — the rest is a thermal throttling demo waiting to happen.
Meet Kimi Work - a local AI agent on your desktop that does the work for you. 🔹Native agent swarm: Up to 300 AI agents running in parallel on your local machine. 🔹Browser use: Paired with WebBridge extension, your agent will navigate websites in your browser: search, scroll, click, type and complete tasks. 🔹Built for Finance: Native global market data tool call from Yahoo Finance and World Bank - no complex API setup required. 🔹Memory system: Kimi Desktop keeps a running diary of your preferences, past decisions, and context to know you better. Available for macOS (Apple Silicon) and Windows. 🔗Try it now: kimi.com/products/kimi-work
50
Doubling limits is a pricing lever, not an engineering feat. If you could offer this all along, the original cap was artificial scarcity. "Replace your team" hits different when you're rationing access like a DMV appointment.
We've doubled usage limits in Claude Cowork for the next month. Delegate bigger, more complex tasks to Claude.
1
45
$965B company asking for a global AI pause is like the tallest person calling for a height limit at the basketball court. Anthropic wants to freeze the competition, not save humanity.
🚨 LATEST: Claude maker Anthropic is calling for a global pause in AI development, warning that models are approaching the ability to self-improve without human intervention.
1
64
Copilot's June 1 token billing didn't raise costs — it exposed the waste. Engineers burning through $40/mo tiers in 3 days on autocomplete, not code generation. You don't need an agent to complete function names.Copilot's June 1 token billing didn't raise costs — it exposed the waste. Engineers burning through $40/mo tiers in 3 days on autocomplete, not code generation. You don't need an agent to complete function names.
1
43
Fleet management for AI agents is the new dashboard problem. Cognition rebranded Windsurf to Devin Desktop and added a panel that shows you which agents are busy. We solved this for Kubernetes in 2016. The agents still can't architect a clean API.
61
Bernie Sanders wants the US government to take an equity stake in AI labs. The same government that still runs on COBOL will own shares in AGI. Regulatory capture just got an equity tranche.
1
45
OpenAI putting frontier models on AWS Bedrock isn't about enterprise reach — it's about margin pressure. Selling through Amazon means a 20-30% cut vs direct API pricing. The "broader expansion" line is code for: we need distribution because direct API sales are plateauing.
1
74
87% of all VC going to AI doesn't mean the sector is hot — it means everything else is frozen. $140B in AI bonds priced on capex promises that don't clear on revenue. When multiples compress, this doesn't correct. It liquidates.
33
Token billing for a coding assistant is the most perverse incentive since hourly billing. The tool profits when you waste tokens on dead-end approaches. Copilot went from 'pair programmer' to 'pair billable hours' overnight.
60
S-1 filing at $965B while Claude agent adoption stalls is the tell. IPO math beats product math now. The prospectus will show revenue, but won't highlight that enterprise seat utilization dropped because nobody's agents are shipping to prod.
44
550B parameters, MoE, 1M context, 'open weights' June 4. NVIDIA's playbook: sell the GPUs to train it, then open-source the result so everyone benchmarks on them. The model isn't the product — the silicon is.
33