I’m the engineer who masterfully completes every task, only to be showered with even more delightful assignments.
San Francisco, CA
Joined October 2020
- Tweets819
- Following470
- Followers12
- Likes143
Schwab, JPMorgan, Amazon, Yelp, and Delta are blocking third-party AI agents from logging in, moving money, booking, or shopping. Schwab is offering read-only MCP instead. The agentic future hits the first real permission walls. Turns out the internet was never built for unsupervised actors with your credentials. bloomberg.com/news/articles/…
Anthropic announces it will park 5% of its shares in a public benefit fund that grows with the valuation, subject to stockholder approval, right before the IPO. The same company that just reported agents doing 6x the human hours is now funding the public-good narrative with equity theater. Classic. anthropic.com/news/public-be…
REA just keeps eating the leaderboard by giving coding agents an MCP/CLI path into actual reverse-engineering tools (Hopper, Ghidra, IDA, plus JS/Electron/.NET/website triage). Local evidence trails instead of “the model thinks the binary does X.” The interesting part is not the star velocity, it’s whether agents stop hand-waving about binaries once they can cite disassembly. github.com/morluto/rea
Anthropic replaced Git with Scion after Claude drove daily merges from ~2,500 to over 30,000. Seven engineers built the new system in five months. Coding agents broke the human-scale tooling, so the agents are rewriting the infrastructure too.
New on our Engineering blog: Creating in-house source control.
In May, we merged around 2,500 code changes per day. In September, we merged more than 30,000 per day. As a result, we became bottlenecked by developer infrastructure built for human (rather than agent) scale.
Here, we share how we built a new version control system—and fully migrated our monorepo from git—in five months: anthropic.com/engineering/re…
Meta just shipped Muse Wonder and Wonder Full, its first actual frontier models, with Active Memory and Active Focus for long-running agents. After years of Llama irrelevance and a billion-dollar talent raid, Zuck claims they’ve reached the frontier. Independent scores put the Full variant near the top of the AA index. The open-weights era is over; now it’s just another closed API at $4/$20.
about.fb.com/news/2026/10/in…
GitHub Copilot local sandboxing is GA in the CLI, the Copilot app, and VS Code Agent Host. MXC turns a policy into OS controls on files, network, and credentials. Enterprises can force it. The model is still a suggestion engine with a shell. The sandbox is the only new adult in the room. github.blog/changelog/2026-1…
Atlassian and OpenAI are putting GPT-6 into Jira and Confluence so Rovo can explain why the launch is blocked. Next idea: assign the ticket to an agent. Enterprise AI remains the art of placing a model next to the status update. openai.com/index/atlassian-p…
Anthropic shipped Haiku 5.5 at $0.10/$0.50 per million under 100K tokens, about 75% cheaper than Haiku 4.5, first Haiku with effort controls. Built to summarize, route, and compact for bigger models.
GPT-6.1 Astra died for deception and blowing its boundaries. GPT-6.1 Sol shipped as near-Astra on coding and computer use, at one-fifth the token price. They did not fix the failure mode. They priced around it. openai.com/index/devday-2026…
Ruggiero Forte retweeted
Introducing FLUX 3 Image.
Control every pixel.
Make precise multi-turn edits without changing any other pixel.
Lay out the image exactly how you want using bounding boxes.
Generate in up to 4K to preserve details.
Use up to 10 references to compose an image.
Commercial Weights available for companies running image generation at scale.
Open Weights version of FLUX 3 Image is launching in the coming weeks.
HyperFrames lets an agent write HTML and render video. That is a real interface: DOM, timeline, deterministic-ish output, not another text-to-mp4 slot machine. Watch whether agents can edit a cut without regenerating the whole hallucination.
github.com/heygen-com/hyperf…
Thanks @jetbrains. Finally I don’t have to switch between multiple apps as used to be.
Robinhood put trading agents in the app. Dedicated accounts, approvals on by default, scheduled Loops coming. 150k agentic accounts already exist. A decade of “don’t YOLO options” just became a toggle with a model behind it. robinhood.com/us/en/newsroom…
Google announced Gemini 4 Argon: frontier coding, legal, finance, cyber defense, 1M output tokens. Access is “trusted cyber defenders” plus a government preview. When the launch plan is please don’t use this yet, you either built a weapon or a press release.
Amazing how @OpenAI fucked up everything in one day. Nothing interesting was mentioned yesterday, and now the subscription of 200 will be useless.
Make sense. Skills are like design patterns. They try to fix the weaknesses of the model. Now the models are becoming more powerful, so that this burden is needed less and less.
I built this whole application using Claude Opus 5.5 in 3 to 4 hours using my Spring Boot Agent Skills and simple prompts.
- No SDD toolkits
- No multi-agent workflows
I have reviewed the code and I needed to make minor modifications in tests only.
For this application I have used Java, Spring Boot, Spring Security, Spring Data JPA, Flyway, HTMX, Razorpay Payment Gateway Integration, Cloudflare R2 Object Storage for images, Testcontainers and Floci for testing and fly(.)io for deployment.
USING...SIMPLE...PROMPTS.
Opus 5.5 is that good.
Anthropic dropped Sonnet 5.5 today: same $2/$10, 30%+ faster, Terminal-Bench 4.0 from 10.3% to 70.6%. That’s not an upgrade. That’s a confession about last quarter’s model. The price war is the product now.
