Michael Preuss

Operator building agentic GTM and AI workflows
Blaxel

[Michael Preuss @mpreussai]

If my market signal agent can't explain why a competitor update matters to our customers or our strategy, I don't see much reason to spend our time and money on it. We can keep watching something without deciding we need to respond.

[Michael Preuss @mpreussai]

I keep seeing recruiters call marketing leaders who can build, work on brand, run paid, and set strategy unicorns. I'm one of those people. I've worked with several others.

We worked hard to develop that range. I wouldn't lower the bar because someone called it a unicorn.

[Michael Preuss @mpreussai]

One thing I had to get sharp on when analyzing a consumption-based business: I needed to see free-credit usage too. Leaving it out loses part of the adoption picture. But I still needed a separate view of paying users to understand what was changing and what to do about it.

[Michael Preuss @mpreussai]

I've started using the gaps in our AI citations to choose what to write. I still have to make it worth reading, but before I spend that time, I want a better reason than needing another topic for the content calendar.

[Michael Preuss @mpreussai]

If I'm asking an agent where to spend time and money, I want it to know who we're trying to reach. Otherwise I could get a very thorough explanation of why we should acquire more people we don't actually want to acquire.

[Michael Preuss @mpreussai]

I expected acquisition costs to rise when I tightened our paid targeting. I was okay with that. We won fewer of the more expensive auctions and spent less overall, which surprised me. Across the paid program, we also spent less per qualified sign-up.

[Michael Preuss @mpreussai]

For Guild, I built a workflow that turns a selected idea from a meeting transcript into a prototype.

I'd rather try it than book another meeting to check whether we imagined the same thing.

https://t.co/iAnJVQHdEo

[Michael Preuss @mpreussai]

One of my fav agents I've built reads Reddit, Hacker News, and competitor sites, then posts its top 3 recommendations in Slack each week. We work through what makes sense for us as a team. What we reject helps us improve the agent too.

[Michael Preuss @mpreussai]

After reading Simon Heaton's post, I built an agent that uses Profound's MCP to review our AI citations each week. I was a marketing team of one, so I wanted help finding where we were missing and deciding what was worth doing about it.

https://t.co/w30EV10LNI

[Michael Preuss @mpreussai]

I started my knowledge graph and Market Signal agent before day one at Blaxel. I arrived with questions to work through with the team.

The first week has enough admin.

https://t.co/iAnJVQHdEo

[Michael Preuss @mpreussai]

I might look a little crazy reading my marketing copy out loud. Probably even more bizarre with the technical content. Luckily, I don't really care.

I catch things I missed reading silently. If it's important, I don't consider it finished until I've read it out loud.

[Michael Preuss @mpreussai]

Building a churn analysis with AI, I needed separate views of total activity and paying-user activity. A drop could mean different things in each.

We could move incredibly fast and adjust as we went, without starting over. That's what I find so cool.

[Michael Preuss @mpreussai]

"Oh my God, I can't believe I thought this was good yesterday."

I've had that reaction to AI-assisted writing I'd already edited critically. Now I leave important pieces until the next day, then reread before sharing. I can keep other work moving while they sit.

[Michael Preuss @mpreussai]

I built a paid-media agent that ran at 5:00am PT, analyzed performance, updated a dashboard and posted to Slack.

The same API and MCP connections let me set up campaigns and make changes. That's a recurring job agents are perfect for.

[Michael Preuss @mpreussai]

Babe, wake up - it's 3/6/9 weeks now. https://t.co/iAnJVQHLtW

via @LinkedIn

[Michael Preuss @mpreussai]

AI was supposed to reduce how much I wrote. I'm writing more marketing and technical marketing content than ever.

Turns out, getting the ideas out of my head and into a draft still leaves a lot of work before it's worth a prospect or customer's time.

[Michael Preuss @mpreussai]

I found roughly 80% of our LinkedIn spend was going to entry-level engineers in Bangalore. Our ICP was senior technical leaders and executives in San Francisco.

I reset the targeting knowing impressions would drop (and they did). I wanted a baseline we could build from.

[Blaxel @blaxelAI]

Not all agent data belongs in the same place.

Some data should disappear with the session. Some should persist for months. Some needs to be shared across agents.

Our blog post explains the tradeoffs and how to choose the right storage option on Blaxel.

https://t.co/9jC2vmrNOJ

[StarSling (YC X25) @starslingdev]

https://t.co/WcstFifXSa

[Blaxel @blaxelAI]

Every harness thinks in files. They all have their sandbox and read and write to its filesystem when they work.

But when two sandboxes need to share context? Teams are back to building S3 export-import pipelines.

We built the solution. Meet Agent Drive.