Hi, I'm Peter Kaminski.
I want to help you use personal AI to succeed in life and in business.
At the dawn of the Internet age (around 1992), I helped people get on this new thing called "the Internet" — phone lines and email and the World Wide Web. I published a popular free guide and worked with individuals and businesses, getting them online and productive with this new global network.
Today, in 2026, I'm helping people work with this new thing called "agentic AI," or — without the jargon — "personal AI": getting them up and running and productive with this new assistive technology.
Think of personal AI as a tool, but a "human-shaped" tool. You work with an individual bot that has a personal name, memories of who you are, and an understanding of what you need to work on and how you like to work. It's long-lasting: it accumulates context about you and your work over months, instead of starting from scratch every conversation the way a chatbot does. And it works inside boundaries you set — free to go ahead on its own for some things, required to check with you for others. You can speak with it out loud, but more often you interact with it through text — chat, instant messages, emails, documents — much the way you text with friends, family, and co-workers online.
Your bot speaks your language — English, Spanish, Chinese, Dutch, Swahili, and dozens of others. It knows how to work your computer — editing and creating documents, using your apps, using the web. It happens to be an expert software developer, too, and can build you pretty much any kind of app you can describe. Using your personal AI to build small, custom apps that do exactly what you need, exactly how you like to do it, is one of the more powerful ways to use AI — and it's available to you.
The technology for all this is brand-new — it came together at the beginning of 2026, so it's still rough and a little challenging to use. But with a little help and guidance, if you can use a computer, you can use personal AI. It's not just for techies — it's for anybody with a little patience, a little curiosity, a little fearlessness, and, most importantly, work that they need done.
What I'm focused on now
A lightweight snapshot, meant to change over time. Last updated: 2026-08-03.
- Giving people a way to start. pkai-agent is a free starter kit: download it, open it in Claude Code, and your own assistant introduces itself and walks you through the rest. It's at v1.2.0 and it's the piece of my work I most want people to actually use.
- Teaching. I ran fourteen free live Open House sessions through July — no slides, we picked the topic in the first five minutes — plus one-on-one coaching. I'm working out what the ongoing version of the open sessions should look like.
- Writing. The Age of Personal AI, a book in progress, and shorter pieces as they come.
- Building the small stuff. Tools that came out of doing this work every day — a collaborative markdown editor, publishing infrastructure, and Swamp, a protocol for people and their agents to talk in public with no platform in between.
- Working out delegation. How much authority you hand an assistant, how you bound it, and how anyone can tell afterward which decisions were yours and which were the agent's. It's the unglamorous problem underneath all of the above; I spent a week on it in June at a human-and-agent unconference run out of law.MIT.edu.
Topic scope
Superb at
- Agentic AI workflow design for knowledge work (planning, execution, supervision).
- Claude Code setup, practice, and pragmatic patterns beyond "coding only."
- Making tool ecosystems legible: surfaces, boundaries, and reliable operating procedures.
Can help a lot with
- Tool selection across Claude Code, Codex, Cursor, Manus, and others—fit to real constraints.
- Lightweight project management artifacts that improve predictability with agents.
- Versioning discipline (docs, repos, file sets) so AI work doesn't sprawl.
Happy to discuss (not expert)
- Org-level governance and policy beyond small-team practice.
- Highly regulated or correctness-critical domains (legal, medical, etc.).
- Deep model internals beyond an operator's working understanding.
Timeline notes
I separate this into observations and a working model. The model is a set of bets, not a promise or a prediction schedule.
Observations
- Practical agentic workflows became widely visible in software development in 2025.
- Capability and adoption accelerated through late 2025 as tools matured.
- By early 2026, non-coders could often use agentic tools effectively with good setup and practice.
My current model
- Agentic AI will continue moving "up the stack" from software into general knowledge work.
- The main bottleneck is less raw capability and more supervision, surfaces, and operating patterns.
- The quality of judgment around agent use will matter as much as the tools themselves.
What would change my mind
- If agentic systems remain fragile outside narrow domains.
- If governance and trust concerns broadly block adoption.
- If new interaction paradigms make today's framing obsolete faster than expected.
If you're looking for ways to work together, start at the homepage .