In June of 2022, I sold my agency Krit. We crossed $1M in annual revenue for the first time in the year before the sale. Ever since we sold, I've wondered - how could I do it faster next time? How would I speed run my way to another $1M agency? Here's how 🧵

Been doing this for MetaMonster for a while, felt like a bad idea at the beginning of 2025, feels like the clear right answer now Cursor is essentially my CMS

There's a B2B lesson in here somewhere

- MetaMonster will automatically identify topics, keywords, and page categories -> which unlocks internal linking and more advanced workflows - Improved integrations so you can deploy all of your updates in one click!

- Easily test workflows on a small subset of pages so you don't burn through your credits - Add workflows to a queue so you don't have to wait on one to finish to queue up the next

You no longer have to wait for a crawl to finish to start queuing up your first workflows. This makes the overall experience of getting started with a new project WAY faster. Next up: - More guidance and education on your first run

We've made some big improvements to the MetaMonster crawling and onboarding experiences over the last month!

If you're ever been in a similar situation or know someone who has I hope it resonates and you find it helpful. https://t.co/kqiIEI6JIz

So I've let it sit and percolate in the back of my head for the last 2.5 years, and last week I sat down to take another crack at it.

As soon as it happened I knew I wanted to write about the experience, both to document how I felt and destigmatize something that had been such a huge fear. But everything was too fresh, and the drafts I wrote weren't coming across the way I wanted them to.

Three years ago I sold my agency to one of our long time clients and went to work for them. I made it 11 months before I got fired.

Every SaaS is super easy to build... ...until you start building it and realize that there are 10,000 edge cases... ...and then even after you've handled all the edge cases, you realize that distribution is actually the hard part

That's another key to prompt engineering: LLMs are probabilistic, so you need to test prompts across a decent-sized dataset to get a real understanding of how well they work.

Maybe it's weighting a certain word heavily, or associating it with something you didn't consider. (at least I think that's what's happening) The new UI makes it easy to generate this at the same time as the output from your prompt inline, and then review them at scale.

While LLMs aren't really thinking, this can help you understand how its model may have "interpreted" your instructions and at the very least give you ideas for what to tweak to get the output closer to what you're looking for.

Something we've found helpful when testing prompts for MetaMonster: have the LLM output it's "rationale."

Pretty cool: - I crawled our site - Told the AI to analyze my page titles and generate an engagement score from 1-10 - Filter to < 5 - Generated optimized titles - Had the AI score the optimized version and provide a justification

Detroit rocks

Anyone else feel like they're racing against the clock in a heist movie every time they go to input their MFA code?

Testing ChatGPT deep research to see how it handles the SEO tasks we're building the MetaMonster agent to be able to handle and I got this gem 😂

LOVE this newsletter CTA from balsamiq