Once your agent has an allowance, the question is what it should spend it on. Verified, validated information — because the same wave of agents that makes budgets necessary is also flooding the web with slop.
Ads can't pay for the internet once agents do the browsing — micropayments will, powered by crypto and new protocols like x402. Your agent will get an allowance, and you'll have to trust it to spend your money wisely.
Agents are becoming the new execution environment — after the OS, the browser, and mobile — and the application layer is moving into agent skills. Instead of writing ad-hoc software, we encode the logic into skills that agents run when needed.
If LLM providers can read every query, the obvious defence is to stop sending them meaning: transform the query into something else, get the answer, transform it back. Call it semantic homomorphism — anonymization is its simplest form.
NVIDIA came out with an initiative to back open weight models, and OpenAI, xAI and Google lined up behind it — against Anthropic's push for safety regulation. What reads as a safety debate is also a fight over who gets to stay at the top.
Your LLM provider knows who you are and sees every query, response, and file you upload. That turns Big Brother from a conspiracy theory into working reality, and we're missing the security controls to push back.
As search shifts from Google to agents, new players like OpenAI, Anthropic, and Perplexity finally have a shot at breaking Google's grip. Good news for the local businesses too scared to even edit their Google profile.
When governments started blocking and pausing frontier model releases, everyone realized how dependent they are on a handful of labs. The geopolitical response points straight at open weights — and a real shot at your own LLM on your own device.
When agents act on our behalf, the web's UX shifts from human-focused to agent-focused. That shift is a big opening for small startups.
As assistants get more capable, the phone dissolves into modular wearables and the web gets rebuilt agent-first, for machines rather than humans.
My 2026 Mac reset, now a single bootstrap script. Also how switching to Claude Code for almost all my coding reshaped the toolset and retired the IDEs.
Why AI agents are a new kind of software rather than hype, and how phones turn into personal assistants that handle real-world tasks. Plus the infrastructure and legal changes it will take.
Five simple principles that boosted every team I worked with: focus on customers, experiment and move fast, simplify, automate, and ensure reproducibility.
How I use Google's NotebookLM as an interactive RAG to skim ML papers, including the Audio Overview prompt I rely on to stay current.
The 2025 run of my from-scratch Mac setup: MacOS preferences, Homebrew, SSH keys, dotfiles, and the tools I install.
A step-by-step log of resetting a MacBook from scratch: MacOS defaults, Homebrew, dotfiles, developer tools, and Tailscale.