Henrik
@hholen.bsky.social
I'm Henrik. I build things and write about them. Products across AI, IoT, and SaaS, including one that got acquired. Writing about the parts of product building that AI changed and the parts it didn't. https://lodd.dev, https://holenventures.substack.com
You should use AI to fake data for your prospects. holenventures.substack.com/p/you-want-y...
The One Place You Want Your AI to Hallucinate
An idea for founders: enterprise software traps its value behind the security gate, and fake data is how you let a prospect feel it first.
holenventures.substack.com
How is it possible the Gmail doesn't have a "book this meeting" button, triggered by an email containing phrases like "let's book this meeting"...
Trying to build a small app using Gemma4 and OpenCode. It is truly mindbogglingly slow, but it's happening. I assume I can fix my setup in some way to reduce it from stopping all the time...
I’m so happy the word “workflow” is now a magic phrase. Thankfully, I never use that term in my day to day... what's that word again... workflow… aww, goddammit.
Henrik Holen just now A year ago, I built a RAG-based AI assistant for architects using n8n. Went deep into it, then AI coding took over. Now I built a marketing workflow with lodd.dev as a source. The visual element in n8n is underrated, great way to co-work with agents. lodd.dev/blog/n8n-mar...
Build a marketing agent in n8n with analytics, web search, and AI visibility testing - Lodd
Build an AI marketing agent in n8n that pulls web analytics via MCP, searches for mentions, tests whether ChatGPT knows you exist, and reports to Slack weekly.
lodd.dev
Adding Google Search console MCP alongside lodd.dev is such a time saver: lodd.dev/blog/connect...
Give your agent more context: Google Search Console + website analytics via MCP - Lodd
Give your AI agent search ranking data alongside website analytics. Connect Google Search Console via MCP in 10 minutes and let your agent correlate queries, traffic, and conversions in one session.
lodd.dev
Anthropic’s acquisition of Stainless is a clear sign that bringing the context into agentic workflows is a priority. Every had a good interview with the founder. Worth having a listen. (I had to run their MCP eval harness on my lodd.dev web analytics MCP. It got a 100%.) every.to/podcast/he-s...
🎧 He’s Building the Plumbing for AI to Use the Internet
Stainless founder Alex Rattray on MCP, a protocol giving LLMs the tools they need to do real work
every.to
Claude lets you build a passable something in a weekend, but where is your moat? Might be good to add some physicality to your ideas. holenventures.substack.com/p/build-some...
Build Something You Can Touch
AI made the software layer free, but the physical world still has moats. Why the most interesting building opportunities moved outside the laptop.
holenventures.substack.com
New post on the lodd.dev blog. Three REST API calls to add analytics context to any automation, for instance, auto-creating a GitHub issue when bounce rate spikes after a deploy. lodd.dev/blog/analyti...
Give your workflows analytics context with 3 REST API calls - Lodd
Add website analytics to any automation with Lodd's REST API. Three calls for traffic, pages, and sources. Example: auto-create GitHub issues when bounce rate spikes.
lodd.dev
One my favourite things I've built is a daily digest of interesting news running on @triggerdev.bsky.social. I put a template for it on github so you can put your interests into it: github.com/hholen/daily...
GitHub - hholen/daily-digest-trigger: AI-powered daily news digest pipeline built on Trigger.dev — collects from HN, YouTube, newsletters, blogs, and Reddit, analyses with Claude, posts to Slack
AI-powered daily news digest pipeline built on Trigger.dev — collects from HN, YouTube, newsletters, blogs, and Reddit, analyses with Claude, posts to Slack - hholen/daily-digest-trigger
github.com
It doesn't matter how much context your agents have, if they don't use it. Right now, the best solution seems to be agent instruction files, but those also need to be kept tight to work. Trying to figure out the best approach for lodd.dev
Anthropic's SWE roles are up 186% since early 2025. But one team's listing says Claude writes most of their code and "the bottleneck isn't typing, it's judgment and taste." Same job title, whole new job.
I wrote a guide to adding Google Search Console as an mcp. Works very nicely alongside lodd.dev. lodd.dev/blog/connect...
Give your agent more context: Google Search Console + website analytics via MCP - Lodd
Give your AI agent search ranking data alongside website analytics. Connect Google Search Console via MCP in 10 minutes and let your agent correlate queries, traffic, and conversions in one session.
lodd.dev
Share updates with your investors, even your angel investors. Don’t let them find out what’s happening in a press release. Just apropos of nothing…
Early Google Docs couldn't do half of what Word could. Janky formatting, no offline mode, complex documents were a mess. But you could share a link and see edits in real time, and it turned out not emailing v7_final_FINAL.docx was worth giving up everything else.
Sometimes, making the experience a little worse make the results a little better. holenventures.substack.com/p/the-fricti...
The Friction That Works Best Is the Friction People Like Least
When seamless becomes invisible.
holenventures.substack.com
If I commented “this is utterly made up” on every LinkedIn post claiming that a “friend or client” experienced “something bad with AI” but that the poster incidentally “can help do it well”, I’d do nothing but comment on Linkedin…
Customer feedback is so important, like reminding you that some people don't work by themselves, so maybe add the option to have multiple users on an account ¯\_(ツ)_/¯
Per-hour cost was never the full price of a person. Brooks named it in 1975: a team of more than one pays a coordination tax. A team of one doesn't. That's the bit most AI-vs-human cost comparisons leave out. holenventures.substack.com/p/the-mythic...
The serverless function limitation on Vercel free tier is slowly becoming my mortal enemy.
I got way more out of AI when I stopped assuming it can't do things. Not because it has no limitations, but because assuming it can't means you never ask. And the answer is "yes, actually" more often than you'd think.
Probably preaching to the choir here but the CLI/MCP unlock is wild. Just asking Claude to pull all my analytics data and cross-reference it with other sources shows me things I'd never have found manually.
There's an experiment where people tap out well-known songs and others try to guess them. Tappers predicted 50% accuracy. Actual: 2.5%. That's what it's like using your own product every day.
The stuff we call "AI-generated" is mostly what professional content always looked like. Broetry, performative vulnerability, "here's what I learned." AI just learned from whatever performed best. We're not noticing a new aesthetic, we're noticing an old one that used to cost more.
I launched lodd.dev today, headless web analytics for agents. Made a launch video and tweaked the website based on the first feedback, then went back and shaved 800ms off each tool call.
I wake up thinking about AI tools. My whole feed is full of people who feel the same. But most people don't care that any of this is possible. The person who would've bought a teleprompter for ten bucks isn't prompting Claude to build one.
