
Founder @And/or Labs; building intelligence products for media, adtech, and emerging tech.
Calgary, Alberta
Joined February 2009
- Tweets151K
- Following1.6K
- Followers3.9K
- Likes8.2K
Wanting to be your own boss is cool and all until you do it and realize that the lived experience is much closer to being an employee of multiple people.
27
So... my company's first cloud services agreement is now sitting with the counterparty's legal team. I guess this is a milestone of sorts?
1
1
162
Guillermo was right that the first Agent37 chart priced Vercel as if the CPU were busy all month. Vercel bills active CPU, plus memory for the time the sandbox is up.
So I ran a real task on an Agent37 sandbox (2 vCPU, 4 GB): cloned pallets/click, installed it and ran the test suite. 2,626 tests passed in 19 seconds at 48% active CPU.
Agent37 charged me $0.000109 for it, one billed minute. The same run at published rates (estimates): Vercel $0.0021 at the CPU I measured, E2B and Daytona $0.0009 each.
That is about 19x cheaper than Vercel and 8x cheaper than E2B or Daytona, on a CPU heavy task.
🤝 Paid partnership
The CEO of Vercel called us out. We claimed Agent37.com was 25x cheaper, and he said we didn't consider active CPU time.
So we fixed the math..
We gave 100 coding agents a sandbox each and had them vibe code an app. Here's the cost per hour at just 5% active CPU:
Vercel: $9.75 (14x)
E2B: $16.56 (24x)
Daytona: $16.56 (24x)
Agent37: $0.68
We're starting a race to the bottom... see you there.
Find a cheaper sandbox and we'll refund the difference.
1
113
Brilliant bit of 5D chess by @beehiiv to buy native ad placements in children’s books.
1
2
137
kinda sucks that bcuz I don’t write like this people now assume that im gettin a ai to write for me wen the truth is dat im a pedant who just really fuckin cares about language
1
120
Neo VM providers won’t see this coming. I’m about to dust off my crusty old MacBook from the days of yore to run my agent swarm.
1
102
Look how many ports this goddamn thing has. I can plug in an external GPU through the Thunderbolt port and run small models locally.
1
22
Vishveshwar Jatain retweeted
article here
modem.ai/blog/pgvector-to-tu…
Either sell agents to people or sell information to agents. Can’t think of a third viable model outside of these lately.
1
1
133
I’ve always been a marketing generalist with a capital M, but within that, organic content strategy is my most developed skill.
I asked an agent, who is loaded up with all my MCP connectors to develop a content strategy for a client—and here’s what I got.
What I can tell you is that this agent already has a working strategy that’s better than 99% of what you see out in the wild.
You can go really far with this if you give your agents the right set of tools to work with because, well, garbage in garbage out.
In this workflow: @AirOpsHQ, @openseo, @sanity_io, @resend, @buffer, @lettercompany.
2
102
One of the few types of software products that will do well in this era are the ones that apply massive infrastructure to precision work.
Today, you can replicate 80% of software products to 80% of their intended function with a few hours and some reverse engineering skills.
What you can’t replicate is software whose intended function simply cannot be performed without—for instance—running the equivalent of your own data center.
For instance, a handful of big adtech companies that handle bidstream data at Internet scale operate on mind-boggling, custom-tuned hardware infrastructure.
Think bypassing the OS to route network packets straight into RAM to shave off a few thousandths of a second, or owning 25,000 physical servers, petabytes of DRAM, and banks of high-endurance SSDs to sustain millions of random writes per second.
These companies have no reason to worry about someone vibe-coding their replacement. Their services are rendered as software, but peel back a few layers and you’ll quickly have your perspective changed.
1
151
A particular thing I’m tired of is magic pill AI solutions that promise auto-marketing, auto-growth, etc.
Marketing is not a tower stacking game, it’s more like chess. Unless you’ve got dumb competitors—in which case—lucky you… there is nothing auto about growth.
Growth is hard and nonlinear. The more magical someone’s “AI books you demos” product seems, the less likely I am to want to evaluate said product.
104
Can anybody at @claudeai or @ClaudeDevs help me with an issue about my startup’s inclusion in the “Claude for Startups” program please?
1
1
131







