Tactical insights for first-time founders to outsmart the burn, the churn & the breakdown.

Hey Founder,

There's a new line item creeping into your cost structure.

It's more volatile than your cloud bill and more addictive than headcount.

You can't cut it out. And oddly, the cheaper it gets, the more of it you seem to use.

I'm talking about intelligence.

Whether you're AI-native or simply adding AI features, it's already becoming part of your burn rate.

Sam Altman recently offered every YC startup $2M in OpenAI credits in exchange for equity.

We've reached a point where founders can literally trade ownership for intelligence.

That deal says something bigger than it first appears. It reveals how quickly AI is becoming a core input to building a company, and why it deserves to be managed as deliberately as cash, people, or runway.

Let's dive in.

When the OpenAI deal dropped, most of the debate focused on whether the credits would lose value as inference gets cheaper.

That misses the more interesting point.

For some AI-native companies, the deal may actually be an arbitrage opportunity.

YC’s standard deal is $500K for 7% equity. OpenAI is offering $2M in API credits through an uncapped SAFE that converts later. The final ownership isn’t fixed, but rough estimates put it near 2% at a $100M valuation.

And “$2M in tokens” is slightly misleading. It’s really $2M of future inference spend.

If inference gets cheaper, the credits don’t become useless. They simply buy more intelligence.

So whether the trade is attractive comes down to one question: how much AI will your company actually consume?

Probably more than you think.

Harvey AI reportedly grew from roughly one trillion tokens a month to 12–13 trillion in about five months. Even a moderately used AI agent can already cost hundreds or thousands per month in model calls before hosting, support, and human oversight.

At that rate, giving up a small slice of equity to fund intelligence starts looking less desperate and more rational. 

The OpenAI deal is really a mirror. Intelligence has become expensive enough that founders are willing to trade ownership for it.

Whether you take that deal or not, the same cost is already creeping into your P&L.

The real thing to manage isn’t only the price per token. It’s how quickly your appetite grows. 

The instinctive response to AI costs is: "Inference keeps getting cheaper. This will sort itself out.”

History suggests otherwise.

When something gets cheaper, we rarely spend less. We use more of it.

Economists call this Jevons paradox: more efficient engines didn't reduce coal consumption, they increased it. 

AI is following the same pattern.

Uber rolled out AI coding tools across thousands of engineers and reportedly exhausted its 2026 AI budget by April. Not because AI was expensive, but because cheaper intelligence created more work worth doing.

At the same time, frontier models aren't getting uniformly cheaper.

Anthropic's Claude Opus 5 costs roughly 2x more per token than Opus 4.8 and can be 3–5x more expensive per completed task because it generates longer responses.

Then there's pricing.

Pylon, a 150-person B2B support platform, saw its Anthropic bill jump from roughly $400K to $1.4M a year after crossing 150 seats and losing subsidised pricing.

The workload didn't explode.

The pricing tier did. 

Cheaper AI doesn't automatically lower your spending.

It expands what you're willing to build.

And ambition is the one cost in your business with no natural ceiling. 

You'd think carefully before hiring ten engineers. Each one has to create more value than they cost.

Most founders don't apply the same thinking to AI.

Software is a fixed subscription. AI isn't.

Every new agent or workflow costs like another "employee" joining the team, except it sends you a usage-based invoice instead of a salary.

If it isn't creating more value than it costs, it isn't leverage.
It's burn.

So stop asking: "Can we add another agent?"
Start asking: "Would we hire another person to do this?"

Once AI feels more like payroll than software, your appetite becomes much easier to manage. 

AI spend rarely explodes overnight. It creeps. Once a quarter, list every AI workflow across your product and operations.

For each one, ask three questions:

Then compare your pricing to your AI costs.

If AI costs scale with usage but your pricing doesn't, your margins shrink every time customers use the product.

Budgets tell you when you've overspent. Unit costs tell you why.

Pick one core unit: per user, ticket, report, or order, and track your AI cost per unit every month.

If that number rises while the outcome stays the same, you don't have growth.
You have a leak.

Then cap autonomous workflows with usage limits and sensible human fallbacks before runaway costs surprise you.

Most founders pay frontier-model prices for work a smaller model can do almost as well.

Customers notice the outcome, not which model produced it.

Not every feature deserves premium inference.

Ask: "Is this our competitive advantage, or just table stakes?"

Run commodity features on the cheapest model that meets your quality bar.

Save your most expensive intelligence for the features customers actually choose you for. 

Most AI bills grow because the same context gets sent over and over again.

These look like engineering optimisations.
They're really margin improvements. 

A founder asked me:

You've accidentally become the operating system.

If you're the only person who sees the client's journey end to end, you're also the only one who can spot problems, identify upsells, and connect one project to the next.

That's why the business depends on you.

The solution isn't to delegate more, it's to systemize the customer journey.

Map the entire client experience - from first contact to renewal - and assign an owner, a standard, and a measurable outcome to every stage.

Then hire around the journey, not the org chart.

Instead of thinking "designer," "PM," or "account manager," think:
Who owns onboarding?
Who owns client outcomes?
Who owns identifying the next opportunity?

Your job is no longer to carry the whole customer story yourself, it's to build a company that can.

The goal isn't to make yourself unnecessary.

It's to make sure clients trust the system, not just the founder.

Send it my way, just hit reply.

A few useful resources this week: 

Before you go…

The founders who treat intelligence like any other resource, measured, optimized, and tied to value, won't just keep costs under control.

They'll build companies that can afford to scale faster than everyone else.

That’s the moat.

See you next Thursday,

— Mariya

Hit reply and let me know. I read every single one (for real).

Hey, I’m Mariya, a startup CFO and founder of FounderFirst. After 10 years working alongside founders at early and growth-stage startups, I know how tough it is to make the right calls when resources are tight and the stakes are high. I started this newsletter to share the practical playbook I wish every founder had from day one, packed with lessons I’ve learned (and mistakes I’ve made) helping teams scale.

Mariya Valeva

Find me on LinkedIn