Overhead

What a new company actually needs set up on day one

A short list, in order, with the expensive-to-reverse decisions marked.

2026-07-24 · Leonard

Setting up a company's software estate is unglamorous, and the cost of getting it wrong is not the setup time. It is the migration two years later. This is the order I would work in, with the decisions that are expensive to reverse flagged.

Identity first, and it is expensive to change

Whatever you choose as the identity provider becomes the thing everything else authenticates against. Changing it later means touching every integration. Decide deliberately, use it for everything from the start, and do not let individual tools accumulate their own separate logins.

Domains and email, also expensive to change

The domain you send mail from accrues reputation over years, so pick it once. Set up sending authentication properly at the start. Retrofitting it after deliverability problems is considerably harder than doing it on day one.

Access by group, never by individual

Create groups by role before there are enough people for it to feel tedious. Granting access to individuals is faster for the first five hires and unmanageable by the twentieth, and untangling it is worse than setting it up.

Billing separation, cheap now and awkward later

Separate accounts or projects for production and everything else. It costs nothing at the start, gives clean cost attribution for free, and limits the blast radius of a mistake.

One documentation location, whichever you like

Most of the common choices work. The failure mode is having three, which is what happens by default when nobody decides. Pick one and record the decision in it.


Identity, domains and access shape everything downstream and are painful to unwind. Most tooling choices are reversible. Spend the deliberation budget on the first group and move quickly through the second.