Model routing is getting a lot of attention because it's the easiest AI cost optimization play to understand, but it's just a small part of a much bigger story. Expensive models for hard stuff, cheaper models for easy stuff. Yup, makes sense. But AI spend is becoming complicated (and high!) enough that companies need an actual strategy around it: - Prompt caching to reuse outputs or portions of prompts - Batch processing to run jobs at more cost-efficient times - Reducing input-token volume before prompts are processed + Budgets, guardrails, better pricing, etc. All of this would've seemed like overkill when AI was just a fun little experiment, but it's not anymore. It's become a legit infrastructure cost that's getting more and more expensive each month. "What model should we use?" is an important question, don't get me wrong. It's just only one of dozens you need to ask. DMs are always open if your cloud or AI bill is getting scary and you want another set of eyes.

Just another day in the life

How I sleep knowing I will not have to use Microsoft Teams tomorrow because my company does not subject its employees to unsafe working conditions

We're hiring AEs in San Francisco! If I were an AE looking for a startup right now, I'd care about two things: 1. Is the product in demand? 2. Is there still room to build? Both of these are absolutely, unequivocally, 100% a yes for these roles. The pitch is simple: AI and cloud spend is exploding, and we help companies pay less without sacrificing performance. We've gone from $0 → $1B GMV in roughly three years, but we still prioritize scrappiness, so the best fits for this role are former founders or people who have built things before. If that sounds like you, DM me. I'd love to talk.

Checking my DMs on LinkedIn dot com

It's 2028. You're running late for a flight. You tell your agent, "Do whatever it takes, I can't miss it." It commandeers 39 Waymos and blocks every entrance for the highway from your location to the terminal, clearing a path for you to speed through without traffic. You make it with four minutes to spare and... the flight is now delayed two hours. (Someone else used the same prompt, and since your agent already took over the Waymos, their agent had to hack the FAA and ground the planes until they could get there on public transportation)

Bulk discounts are awesome if you'll actually use everything you bought. The problem is, most don't. Pump.co gets you 15% off AI spend without making you bet on what your usage will look like a year from now. DMs are open if you're curious how it works!

LinkedIn influencer: "A circle looks at a square and sees a badly made circle. A square looks at a circle and sees a badly made square. Most people spend their lives optimizing for a shape they never chose." Chipotle worker: "I asked burrito or bowl."

easiest way to let the world know you're not worth investing in:

RIP Sisyphus. You would have loved personal AI agents. 🙏

I love this image. It does such a great job of explaining why we shouldn't judge others. Like, just because this guy doesn't realize that animals don't understand English doesn't mean he isn't smart in other ways.

saying moat in SF and moat on an ancient battlefield are two wildly different things

Wow, they finally have a visual depiction of how it feels to generate shareholder value:

If your personal AI agents aren't hiring their own personal AI agents who want to hire their own personal AI agents, you're ngmi

Life after you prompt Codex to build you a billion dollar business and you remember to end it with "and please make zero mistakes"

Either A/B testing doesn't work or I'm doing something wrong

Tiny Speck turning their failing video game company into Slack:

Joining a meeting when you work at a company that uses Microsoft Teams:

the NFL equivalent of having 191 agents and $191 of MRR:

Ship of Theseus (2026): if you ship vibe-coded slop that replaces every original feature, redesign the UI, change the pricing, and start calling yourself "AI-native," is it still the same product?