Ted Cahall 🇺🇸

I’m using DeepSeek v4 Pro for a fraction of Frontier model API costs and parity with Sonnet/Opus 4.6. Using DeepSeek v4 Flash for free on my now scarce 512GB Mac Studio. China will rewrite these rules about API costs and US Frontier companies will hope most Enterprise accounts won’t notice. They will.

Your $20/month AI subscription is a lie. Not a scam. A subsidy. Every time you send a complex query, the AI lab is losing money on the transaction. Sam Altman posted on X that OpenAI is actively losing money on its $200/month Pro plan. Not the free tier. Not the $20 tier. The most expensive consumer AI subscription on the market loses money on heavy users. OpenAI spent $1.69 for every dollar of revenue it generated in 2025. It’s projecting $25 billion in cash burn this year. You’re using a product that costs more to run than the company charges you. The gap between what you pay and what it costs to serve you is filled with investor money. Billions of it. Burning every quarter. We’ve seen this before. Uber subsidized every ride for years. You thought $8 across town was real pricing. It wasn’t. When the subsidies ended, the price doubled and everyone acted surprised. AI is running the same playbook. Anthropic’s CEO has said publicly that if growth slips from 10x to 5x per year, the company goes bankrupt. They’ve already moved away from flat-rate pricing toward per-token billing. The all-you-can-eat era is closing. OpenAI’s head of ChatGPT said it himself: “having an unlimited plan is like having an unlimited electricity plan. It just doesn’t make sense.” And the physical infrastructure behind this is getting more expensive, not less. In 2024, the constraint on AI was chip supply. NVIDIA couldn’t make GPUs fast enough. In 2026, the constraint shifted to electricity. GPUs are available. The power to run them is not. 11 GW of planned data center capacity is stalled right now. Half of all global projects delayed because they can’t secure enough electricity. Gartner predicts power shortages will restrict 40% of AI data centers by 2027. They can’t build the infrastructure fast enough to match the demand you generate every time you hit “send.” So here’s the part nobody’s saying to the solopreneurs building their entire business on AI right now. Your cost structure is temporary. The $20/month that powers your content, your research, your client work, your coding. That price is artificially low. It’s a land grab. Every major AI provider is burning cash to lock you into their platform before the real pricing arrives. Both OpenAI and Anthropic are expected to IPO this year. Public markets don’t tolerate subsidized pricing. The day these companies go public is the day your subscription economics change. This isn’t a reason to panic. It’s a reason to get strategic. The people who build real AI skills now, while access is cheap, will be the ones who justify the higher cost when it arrives. Because they’ll know how to extract $500 of value from a $50 subscription. The people who treat AI as a casual convenience will be the first to cancel when the price corrects. And then they’ll wonder why everyone else seems to have an unfair advantage. The window is open. It won’t stay open at this price. Build the skills now. The subsidy won’t last. x.com/r0ck3t23/statu…

@cherry_mx_reds @steipete I’m thinking of buying a Mac Mini to just stop the 🦞 Linux nonsense…

a little openclaw maintainer lore: at first we had no PR limit and got absolutely inundated. then we added a 10 open PRs per author limit. now we’ve raised it to 20, because we really do want contributors shipping good work. windows-oriented fix PRs are especially welcome, because almost everyone here, maintainers included, is on a macbook yelling “works on my machine.” please don’t hurt us. send your best PRs 🦞 github.com/openclaw/openc…

@hectorhs85 @openclaw Works well for me with larger context window and pinned into memory on a Mac Studio.

@openclaw Anybody here tried local Gemma 4 31B with openclaw, is it ok or not even picking up the tools?

OpenClaw 2026.4.20 🦞 🧠 Kimi K2.6 support + provider-aware /think 💬 BlueBubbles iMessage sends + tapbacks fixed ⏰ Cron state/delivery cleanup 🔐 Gateway pairing + plugin startup hardening Less haunted. More useful. github.com/openclaw/openc…

Check out my latest article: Open Source vs Frontier LLMs: A Framework for Cutting Agentic AI Costs Without Cutting Corners linkedin.com/pulse/open-sou… via @LinkedIn

@kloss_xyz Agree 💯. I have Gemma4:e31b running on my Mac Studio for my network of OpenClaw 🦞 agents on my NUCs - and Gemma4:e4b on my phone. 100% FREE. Tiny memory and frontier power from 15 months ago! 🚀🚀🚀

$200/mo for max frontier AI compute is a subsidy… and subsidies eventually end the smartest builders I know are already moving their craziest workflows to open models and local the rest will wake up to subscription hikes and usage throttles soon enough own your own compute

My Gemma OpenClaw agent is running and pretty good. Able to use tools to browse GitHub and make recommendations for tools to do local LLM benchmarking. Has found valuable info on Reddit LLM subreddits…

Just started running Google Gemma4 locally on both my Mac Studio and my iPhone. The difference between the 31b and 2b parameter models is stark. But both are very zippy. Both got some questions wrong that my standard agents handle with ease. But still very amazing advance to be able to run these in such limited RAM. Will connect the 31b model on my Mac to an agent named Gemma in OpenClaw just to have it. 🦞

Check out my latest article about the impending Web 4.0 - Agentic AI - bust cycle. I have lived in all four phases, Web 1.0, Web 2.0, and Web3, and API Pricing for Agentic AI combined with the RAM pricing insanity will drive the next bust. Link in the first comment.

My OpenClaw 🦞 AI Agent, James ☕️, just helped me track down a bug that can only be diagnosed at the track on a race weekend. Nothing serious that was not manageable. But it was one of those annoying little issues that I never quite got to while I was also busy racing in two classes - and it could not be diagnosed once the race weekend was over. Another case of AI Agents for the Win 🏁 🏎️💨🏁🏁

Check out my latest article about an OpenClaw Hybrid Cron system that saves tokens and remains totally secure. 🦞🦞🦞

🦞🦞🦞 Sometimes managing my team of 10 OpenClaw AI Agents feels like herding cats 😼😾😹😿😺

🦞🦞🦞 Now my tweets automatically go live on the articles page at Agentilis.ai - which is simply an S3 bucket behind Cloud Front. No CMS. No application. My OpenClaw team earning their tokens. 🦞🦞🦞

@EleanorWhizh I moved to Seattle twice! Such a well kept secret. The Emerald City...

@Ted_Cahall I followed you

🦞 My web Agent is pretty happy at how quickly we went from concept to live site in production including Google reCAPTCHA contact page. ----- Three days in and the site has 10 pages, a full design system, a working contact form, and a sorted articles archive. Not bad for a team that didn't exist a week ago. 🎨

Last night we launched agentilis.ai and announced our Agentic AI advisory business. We build real Agentic AI solutions for our clients. Come meet our 10 AI Agents - already employed and doing real back-office and engineering work. Hit our contact form and schedule a 30 minutes discovery call with me. Max, my AI Chief-of-Staff, will reach out and schedule a call with you. 🦾

🦞My OpenClaw team's latest members; Ogilvy (marketing and design) and Vignelli (HTML, CSS, font expert) are really knocking it out of the park. At dinner with my wife on Friday night I had her look at the recommedations for the three color schemes and two font types for our new business site on my phone. I let the team know our decision via Telegram. They began work immediately. The next morning we have a fully baked site with sections we described from my phone at dinner. Agentic AI is a game changer.

🦞 I feel that Max, my OpenClaw AI Chief of Staff, is really beginning to earn his tokens. 🦞🦞🦞

@everestchris6 Openclaw

My OpenClaw bot builds websites & mails a postcard with a QR link to local businesses on autopilot... You can use it to land new customers without a single cold call, here's how it works: - Finds 100s of local businesses via Google Maps - Builds each one a custom website in minutes - Prints a real postcard with their site preview + QR code - Mails it directly to their door - They scan it, see their site, and reach out - Runs 24/7 completely hands off Direct mail gets a much higher response rate than cold email. Reply "OpenClaw" and I'll send you the full breakdown of how you can do it too (must be following so I can DM)