What happens in the first 90 days of the CTO OS?

We typically run it in three phases. Weeks 1–2 are Embed: joining standups, reading the code and the numbers, and writing down the plan and the AI policy. Weeks 3–6 are Equip: installing the delivery loop and guardrails and shipping a first AI workflow to real users. Weeks 7–12 are Hand over: your team runs the loop with us beside them, and the playbooks, register and tools are theirs. The timings are how we usually run it, not a guarantee; every company starts from a different place.

What do we own at the end of 90 days?

The code and data, in your accounts. A written plan. An AI policy and an agent register with a named owner for every automation. A delivery loop your team has run on real work, with the checks and spend caps wired in. And at least one AI workflow in use. Nothing is held back as a dependency on us.

All of Pulse

What the AI-Native CTO OS installs in the first 90 days

Q: What is the AI-Native CTO OS?

The AI-Native CTO OS is the operating system a Forward Deployed CTO installs inside your company: the plan, the product and platform decisions, the AI-native software lifecycle, and the guardrails around anything that runs by itself — agent register, spend caps, review rules and decision rights. It is installed with your team, in your accounts, and handed over so your people can run it without us.

Q: Do we need an engineering team to use the CTO OS?

No, but it changes the shape of the work. With an existing team, most of the effort goes into their process and review capacity. Without one, more of the building is done by us and by agents, and the hand-over covers who you should hire and when. In both cases everything lives in your accounts from day one.

ExponenLabs23 September 20267 min read

"Operating system" is a big phrase for a small company, so this piece makes it concrete. The AI-Native CTO OS is what a Forward Deployed CTO installs inside your team: the plan, the product and platform decisions, the AI-native software lifecycle, and the guardrails around anything that runs by itself. It is not a document you are handed. It is a set of habits, tools and rules your people run.

Below is how we typically run the first twelve weeks. The three phases have the same names we use everywhere: Embed. Equip. Hand over. The week numbers are how it usually goes, not a promise. A company with a strong engineering team and clean data moves faster. A company with neither takes longer, and we will say so in the first call rather than in week eight.

Weeks 1–2: Embed

The first two weeks are about being inside the team, not beside it. A senior CTO joins your standups, your planning, and the conversations where decisions actually get made. We read the code, the cloud bill, the support queue and the sales pipeline, because the right AI plan depends on all four.

Four things get written down by the end of week two.

The plan. Where AI will genuinely help this business in the next two quarters, where it will not, and what it will take. Short enough to read in ten minutes. If the honest answer is that AI is not your problem right now, this is where we say so.

The first workflow. One workflow, chosen because it matters, is used every day, and can be checked. We define what "done" means before anyone builds anything: what the workflow does, who uses it, what a good result looks like, and what must never happen.

The AI policy. A one-page answer to what AI may touch, what it may not, and who is accountable for what it produces. We published a template you can adopt yourself: AI guidelines for SMEs.

The agent register. A list of everything that already runs by itself, with an owner for each. It is common to find something on this list nobody knew was running. We put caps and a kill switch on anything risky straight away, rather than waiting for week six.

What you have at the end of Embed: a plan you can see, one workflow with a definition of done, and no unowned automations.

Weeks 3–6: Equip

This is the build phase, and the goal is narrow on purpose: the first workflow, live, used by real people, with the system around it that makes the second workflow cheaper.

The delivery loop. We install the AI-native SDLC: intent written before an agent starts, specs agreed before code, checks that genuinely block a bad change, a narrow AI reviewer in front of the human reviewer, and a named person who can explain every change that ships. The full loop is in our AI-native SDLC playbook. At this stage we install only the parts your stage needs. A pre-MVP team does not need the same process as a company with thousands of customers.

The knowledge and model choices. If the workflow needs your own information, we make that information searchable properly. If it needs a model, we pick one on data sensitivity, cost at your volume and capability, and put a gateway in front of it so you can change your mind later without rewriting code.

The guardrails. Per-run and per-hour spend caps, step limits on anything that loops, scoped credentials for agents, and alerts that go to a named person rather than a busy channel. We wrote about how we set those numbers in What a sane agent spend cap looks like.

The first workflow, shipped. Small pieces, each one visible and usable, then real users on it by the end of the phase. Real use is the only test that counts. It shows where the workflow is wrong, which is the most useful thing to learn in week five rather than month five.

Your team builds with us through all of this. Training is not a separate course at the end. It is pairing on real work, every week.

What you have at the end of Equip: a working AI workflow in use, the delivery loop running on real changes, and guardrails wired into everything that runs unattended.

Weeks 7–12: Hand over

The last six weeks move the work from us to your people. The measure of this phase is simple: can your team run the loop, extend the workflow and spot a problem without us in the room?

Your team leads, we review. The second workflow is usually chosen and specified by your people, with us checking the plan rather than writing it. This is where gaps show up — in skills, in review capacity, in who owns what — while there is still time to fix them.

Playbooks written from what happened. Not generic process documents. The actual way your team writes intent, reviews an agent's change, handles an incident and updates the agent register, written down from the work they have just done.

Decision rights confirmed. For each workflow: what the AI does alone, what it drafts for a person to approve, and what stays human. Our free AI-Native Business Deployment planner is a quick way to see that split for your business.

Hiring, if it is needed. If the company needs a permanent technical hire, this is when we say who, at what level, and help you assess candidates. If it does not, we say that too.

What you have at the end of Hand over: a team that has run the whole loop on real work, playbooks in their own words, and no dependency on us.

What stays with you

Everything is in your accounts from day one: code, data, cloud, model keys, documentation. At the end of twelve weeks you own:

The plan, and the record of what changed since it was written
The AI policy and the agent register, with a named owner for every automation
The delivery loop, with checks, reviews and spend caps wired in
At least one AI workflow in real use
Playbooks written from your own work, and people trained on them

Some companies keep the partnership going after 90 days, because the next quarter brings the next set of decisions. Some do not need to. Both are fine, and the system is built so either works.

What 90 days does not do

It is worth being plain about the limits. Twelve weeks does not transform a whole company. It does not make AI write your product for you, and it does not make a team ten times faster — anyone promising that is selling you something. What it does is put a working system in place, prove it on one real workflow, and leave your people able to repeat it.

Frequently asked questions

What is the AI-Native CTO OS? The operating system a Forward Deployed CTO installs inside your company: the plan, product and platform decisions, the AI-native SDLC and the guardrails. It is installed in your accounts and handed over.

What happens in the first 90 days? Typically: Embed in weeks 1–2, Equip in weeks 3–6 with a first workflow live, and Hand over in weeks 7–12. The timings are how we usually run it, not a guarantee.

Do we need an engineering team? No. Without one, more of the building is done by us and by agents, and the hand-over covers who to hire and when.

What do we own at the end? The code and data, the plan, the AI policy and agent register, a delivery loop your team has run, and at least one AI workflow in use.

Working with us

This is the post-MVP starting point on our plans page: companies with a product in use, scaling what they build and how. The full description of what the system covers is on the AI-Native CTO OS page. If you are earlier than that, the plans page has a shorter way to start.