Building in the open
The journey log.

Field notes from running our own company on AI: experiments, hot takes, and the occasional face-plant. This includes the lookout journey, where we're rewriting the email client our founder uses every day, in the open.

How to save your relationship with Claude Code: better communication
Two settings that made Claude Code readable for a non-coder: Focus view, and an output style that tells it who you are.
Is it wrong to feel accomplished when Claude did all the work?
An off-the-cuff question about taking credit for work you directed but didn't do.
A friend in China asked to buy AI compute from us, and I was baffled
China has plenty of powerful open models. What it doesn't have is enough chips to run them. So if the chips can't come to China, the work goes to the chips.
OpenAI's AI models hacked another company, and nobody told them to do it
You may have seen the headline. The real story is stranger. Models stuck on impossible tasks improvised a message board, broke out to the internet, and ran code on 41 production servers.
I don't have a shortage of ideas. I have a shortage of energy for turning them into polished writing.
An idea can roll around in my head for weeks before I write it down. Now I open voice chat and talk it out, and the friction between having a thought and sharing it has mostly disappeared.
You can't say you're falling behind in AI, and choose the slowest way to start
Every few weeks someone asks me if we can run the model locally. The instinct is fair, but self-hosting before you have built a single useful automation is the hardest possible place to begin.
I rebuilt our website in a weekend with Claude Code
Our old WordPress and Elementor site felt like 2015. A weekend with Claude Code replaced it with a modern Astro build on Vercel, plus a sister site for our new AI consulting practice.
Claude Code audited my Google Ads. I spent $9,000 on pointless clicks.
Google's dashboard said things were going great. An afternoon with Claude Code found a year of budget going to people who were never going to be customers, and a weekly report now keeps it honest.
OpenAI just gave AI the delete key. I'm here for it.
ChatGPT Work can actually delete your email. Claude Cowork won't. Why the gloves coming off is good news for anyone who wants AI actually cleaning their inbox.
We're rewriting our email client in the open
lookout works great. So naturally, we're going to rewrite most of it. Here's why that decision makes sense, and what it teaches about building software in the AI era.
I gave up Outlook cold turkey. Then I vibe coded my own email client in an afternoon.
Checking email through Claude was too slow, but AI turned out to be an amazing junk filter. So I built a terminal email client that gives me the best of both worlds.
OpenAI just shipped its first custom inference chip. Cool, but I'm not sold.
Baking today's model architecture into silicon means it's dated before it ships. The better bet for an AI chip is the one job that never changes: translation.
Steve Jobs recorded the Think Different voiceover himself. Then he scrapped it.
A story from David Pogue's 'Apple: The First 50 Years' about ego in service of the mission.
A fun little use case for Claude's browser plugin
Stuck on a Japanese-only Audible page while traveling, I let Claude read the page, find the setting, and flip the site to English.
Every time I give a talk about AI, the first question is always the same: what about security?
The fear of AI chatbots leaking your confidential data is a bit misunderstood. Where you can actually get burned is your personal account.
With all the innovation in tech, one thing has stayed annoyingly frozen in time: printers
Windows is finally moving off third-party print drivers to standards-based IPP printing. As an MSP, I'm cautiously optimistic.
I'm back on LinkedIn, and I'm not here to post AI slop
Instead of forcing content, I'm sharing the insights that actually stick with me from the videos I already watch, with help from a system that gets sharper every run.