What should you build, and what should you buy?
The architecture practice · Checked 20 Oct 2026
We decide build or buy one block of your plan at a time, never for the whole program. Where a licensed product does the job, you buy it and we connect it to the board. Where the work is specific to your business, or crosses systems no single vendor can see, you build it and you own it.
Build only the parts that carry your advantage. For every block you buy, we name one product, with the reason and the trade-off written next to it.
Chapter 1 · one block at a timeWe decide each block of the plan on its own
A program-wide answer is almost always wrong, because most plans mix commodity parts with work only you do. So we draw the plan first, then mark each block build, buy, or keep.
fig 1 · build-or-buyChapter 2 · four questionsFour questions decide every block
| We ask | If the answer is yes |
|---|---|
| Does a licensed product already do this well? | Buy it, and connect it to the board. |
| Does the work cross systems no single vendor can see? | Build it. No vendor will build it for you. |
| Is the decision specific to how your business competes? | Build it, and own it. |
| Would a written rule or a report answer it? | Build the rule or the report, not an agent. |
Chapter 3 · what we build, and what we don'tWe build what crosses systems and carries your advantage
| We build | We don't build |
|---|---|
| Work that crosses systems no single vendor can see | Commodity parts that a licensed product already does well |
| Decisions that are specific to your business | An agent inside one vendor's product, where theirs is usually better and included |
| The parts that carry your advantage, so you own them | An agent for work that a written rule or a report already answers |
A vendor's own agent stays inside the vendor's product. It joins the board as a client like any other, under its own identity, and writes to the same record.
Chapter 4 · one recommendation per blockEvery block you buy gets one written recommendation
We keep a short list for each block, not every product on the market. Then we pick one and write down why, what it costs you, and when to look again.
block: e-signature for policy documents decision: buy shortlist: 3 products, scored on the same 6 criteria pick: product B why: already licensed in two of your entities, and meets EU residency trade_off: priced per envelope, which costs more above 40,000 a year revisit: at renewal, or when volume passes that line decided_by: your enterprise architect, 2026-09-18
Chapter 5 · costs to price firstWe price the licenses before the pilot, not after it
Around three dozen subscribed applications is normal. Each one's rate limits and seat pricing decide how much an agent can do in it, so we check them before we design the work.
| The cost that surprises | What we check |
|---|---|
| Seats for agents | Whether an agent working inside the product needs a seat of its own. |
| API rate limits | How many calls an hour the plan allows, against the volume the job needs. |
| Per-call pricing | What the job costs at full volume, not at pilot volume. |
| AI spend you've already committed | If your agreement covers the runtime, we use it, and we don't bill for it. |
Chapter 6 · when a decision changesA build or buy decision is reopened on a date
- At renewal. Every bought block comes up again before its contract renews.
- At a volume line. The trade-off we wrote down says when the price stops working.
- When the vendor ships it. A part we built can move to a product once the product does it well.
- When it becomes your advantage. A bought block that starts to set you apart is a candidate to build.
Questions architects ask
Will you push us to build more?
No. Every block you buy is a block we don't have to build, run, and answer for. If a product does it well, we say so.
Who owns what you build?
You do: the code, the documentation, and the record of every decision behind it.
What if we've already bought a product for that block?
We start from it. A licensed product that does the job stays, and we connect it to the board.
How long does the plan take?
We draw your plan in five days and mark each block build, buy, or keep. You keep the plan either way.