Tactical insights for first-time founders to outsmart the burn, the churn & the breakdown.
Hey Founder,
If you run an agency or services-heavy business, you’ve probably felt the shift.
Traditional agencies scale with headcount and sit around 20–30% margins. Whereas the new AI-native stories point to much leaner teams and materially higher margins.
But most founders are getting it wrong.
They are keeping the same delivery model and layering AI on top: same headcount dependency, same linear scaling, same custom work, just faster.
That’s not AI-native, it’s an upgraded agency.
This issue is about what actually needs to change to reach software-like margins, and how to tell if you’re building a system or just speeding up old workflows.
Let’s dive in.
The Margin
The Six Dollars Everyone Is Chasing
Sequoia’s framing is simple: for every $1 spent on software, companies spend ~$6 on services to actually get the work done.
If you run an agency, you know exactly where that $6 goes: into people doing the work.
That’s why software kept software margins, and agencies didn’t.
AI is starting to change that split, not by removing services, but by changing who captures that $6.
What’s shifting is where the “intelligence” lives.
Work that used to require teams: research, drafting, QA, and analysis, is increasingly handled by systems. Humans stay where it matters: judgment, edge cases, and client relationships.
You can see it already:
Outbound used to be tools + SDR teams, now systems like 11x.ai can find leads, write, follow up, humans just supervise.
Support is moving the same way: full conversations handled end-to-end, with humans stepping in only when something breaks. Companies like Fin AI are leading that change.
Similarly, Harvey is changing how legal work looks. Harvey handles the drafting, reviewing, and answering queries across workflows.
The shift: from assisting lawyers to executing the work itself.
Other examples making the shift from assisting humans to actually doing the work end-to-end:
The client still pays for the outcome, but the $6 is captured by whoever owns the system that delivers it with far less human effort.
That’s how a service P&L starts to shift: less headcount, faster cycles, more consistency, and margins that actually improve over time.
So the real question isn’t “are services back?”
It’s: what has to be structurally true for a service business to earn software-like margins?
Tiny Reframe
AI-native agencies are secretly data companies
A lot of founders hear “AI-native” and think it means using ChatGPT internally, automating a few steps, or building better prompts.
That’s AI-assisted; you might get productivity gains, but you’re still running the same business.
AI‑native starts from a first principles question:
“If we designed this service today, knowing AI exists, what would the workflow actually look like?”
Once you answer that honestly, the structure changes.
Data becomes the core. Every engagement generates inputs, outputs, and corrections that get captured and reused.
Work stops being bespoke, most of the delivery follows a repeatable, systemised path.
And AI becomes the default executor, handling the repeatable work, while humans focus on judgment, edge cases, and client trust.
The economics shift with it: more revenue per operator, higher margins, and the ability to scale without hiring in lockstep.
Internally, the question changes too.
It’s no longer “who should do this?”
But “what runs this, and where does a human actually need to step in?”
That’s why the cleanest way to describe an AI-native agency is this: it’s a data company that happens to sell services.
If that’s not how your delivery works, you’re not AI-native, you’re AI-enabled.
Both can work; only one changes your margin profile.
Tiny Filter
Should You Even Build an AI-Native Agency?
This is the part most people skip.
Despite the hype, this model doesn’t fit every business.
It tends to break when:
Customers need to own the workflow themselves because of compliance, risk, or internal sensitivity.
It also struggles when there’s no tight feedback loop: outcomes are fuzzy, cycles are long, or data is limited.
And there’s a simpler signal: if your best customers keep asking for the engine instead of the service, you’re not building an agency, you’re building a product.
When AI-native is the obvious answer:
High-volume, repeatable work.
Clear success metrics.
Digital inputs and outputs you can track and reuse.
If most of that sounds like your business, you’re likely in AI-native agency territory.
If not, you’re probably better off going product, or at least hybrid.
Margin Moves to Actually Build Toward AI-Native Margins
1) Re-map one workflow end-to-end
Don’t try to “AI-ify” your whole agency; start with one workflow and make it behave like a system, not a team.
Pick something digital, repeatable, high-volume, and with a clear definition of “done well” (e.g. meetings booked, tickets resolved, invoices collected).
Then redesign it assuming AI handles most of the execution: research, drafting, scoring, first-pass output. Humans step in for judgment and edge cases.
If multiple people still need to touch every job, nothing has really changed; you’ve just layered AI on top.
2. Treat your historical work like training data
Most agencies are sitting on valuable data and treating it like clutter.
Your past work, what you performed, what failed, what needed revision, is exactly what teaches your system what “good” looks like.
The moat isn’t the model; it’s the workflow and the data layer around it.
3. Build a small internal system first
Don’t start with a product; start with an internal system that can run real jobs.
AI proposes or scores the work, a human reviews it, and the system logs what happened and the outcome, then runs actual client work through it.
Track what changes, time saved, output quality, and revisions.
If you’re not materially reducing human time on repeat work, the workflow is still too human-first.
4. Make every job feed a loop
Every job should generate a reusable signal.
Log inputs, outputs, edits, and results, and use your best work to evaluate new work.
That’s how human effort concentrates on the 10–20% that need judgment, while the rest compounds.
5) Productize from the margin backwards
Start with the margin you want, not the service you sell.
What would need to be true to reach 50–70% gross margin?
Usually: tighter scope, cleaner inputs, faster feedback, and more system-led delivery.
Then design the offer around that, sell outcomes, not hours, and see if margins actually move.
If they don’t, your model is still human-first.
Tough Love Corner
A founder asked me this week:
“We just closed our Seed and have our first board meeting in 4 weeks. Never done this before, any tips?”
Congrats first of all!
The biggest mistake here is overthinking it or trying to perform; that’s not what this meeting is for.
First thing I’d do: lock in all four board meetings for the year now, same day, same time. It removes a lot of friction later and makes you look organised from day one.
Then keep the structure simple: quick update, numbers, and one or two real topics that actually need discussion. Not ten updates, just where you want input.
On prep, don’t overcomplicate it: send a rough agenda a couple of weeks before, tighten things a week out, and share everything a few days before with a clear focus. That alone makes the meeting better.
And honestly, the main thing: your first board meeting isn’t about impressing anyone, it’s about setting the tone.
Clear thinking, no theatre, actually using your board instead of performing for them.
If you get that right, the rest follows.
Got a burning founder question?
Send it my way, just hit reply.
Founder’s Toolbox
Reads this week:
Before you go…
You don’t get AI-native margins by using better tools. You get them by changing how the work actually happens.
The winners won’t be the agencies with the best AI branding. They will be the ones who own the system, capture the data, and design delivery so humans step in only where judgment matters.
That’s the real shift.
That’s the real moat.
See you next Thursday,
— Mariya
What did you think of today’s issue?
Hit reply and let me know. I read every single one (for real).
About me
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.
