We put a lot of effort into making it the best private keyboard with voice typing! Works offline, no cloud.

A few weeks ago I said software built for agents will be the next big thing after SaaS. This week Walmart opened its store to AI shopping agents, with spending caps and approvals per agent. Amazon is blocking them. I think this is the paradigm coming for every business: Some will make it easy for agents to buy from them. Some will try to keep agents out. If your customers start sending their agents to order, quote or book, can your systems handle that? Or does it all still need a human clicking through a website?

I think "software is dead" is the wrong conclusion. What I see: companies suddenly want waaaay more custom software than before. Internal tools, automations, knowledge bases, agents that use their existing systems. What changed is who builds it. Small teams who know how to steer coding agents. That's why I started @MadeByAgents. The honest part: most of what gets vibe-coded today is a powerful starting point, not a finished product. Getting it production-ready is still work, usually a ping-pong between business, design, engineering and user feedback.

Asking Claude to review code again and again until it stops finding bugs doesn't work. It never stops. You need a definition of done before the agent even starts. What that can look like: 1. The real user flow works in the running app. Not just "tests pass." 2. The agent is not allowed to edit the tests it is graded on. 3. Low-risk changes get agent review + CI. Auth, payments and data migrations get human review, line by line. 4. Another review round only happens if the last one found something that matters. I think this is the real lesson from Anthropic's post about Claude writing 90% of Claude Code. What's actually got them there is the review process. To decide what needs human review, I would use one of the decision models like Jev or an OS alternative.