Generic on OSS,
the specific with AI.

Customers themselves come to build, paired with AI. But the first move is not writing code. 1-04 showed that a builder does not have to be on the company's payroll. If you can stand on the building side in dialogue with AI, the customer becomes the builder. And the customer holds the business context from the start, so what the SIer used to gather from the outside and translate can be skipped. This chapter looks at where the customer starts.

What you want to build splits into three

It helps to split what you want to build into three: the generic, the personal, and the organization's foundation. The first move differs for each of the three. We take them in turn.

Use OSS for the generic, because that is the most economical

Authentication, documents, the database, video conferencing, the web. These generic capabilities are already shared with the world as OSS. Someone built them, released them, and tens of thousands of people hardened them. There is no need to write them from scratch. There is no need to buy them from a vendor either. Just use them. That is the most economical path.

It is not only economical. It is also effective as a security measure. Widely used OSS is exposed to eyes all over the world. Countless vulnerabilities have been found and closed in it. It is far more battle-tested than code you wrote from scratch. It is far more battle-tested than a vendor product whose insides you cannot see.

There is an inversion here that is easy to miss. The attention goes to AI. But most of the generic work is actually carried by shared OSS. The effect of OSS is greater than the effect of AI. AI is only the tool that quickly builds the specific part sitting on top.

For the personal, add yourself onto OSS in dialogue with AI

When an individual builds a tool for themselves, using OSS as it comes is sometimes not enough. So you take OSS as the base and customize it, in dialogue with AI, into a shape that fits you. There is no need to learn a framework. Say what you want in ordinary words, try what comes back, and ask again. That is all. What you learn is only how to put into words what you want.

This became possible because the cost of learning fell by orders of magnitude. It used to take months to memorize the syntax, get used to a framework, and get one thing running. Paired with AI, having something that runs takes hours to days. What you gave up on when the unit was months, you try when the unit is days. That line has been crossed.

The worked examples are handled by the Independence part.

• 2-04 — write the processing in Python (Excel macros move to Python too)
• 2-10 — build a website in dialogue with AI
• 2-14 — from electronics to IoT

Each of these is about an individual who is not a professional programmer building their own tool with OSS and AI.

For an organization, first stand up the generic foundation on OSS

For an organization, the starting point is the same. First, stand up the generic foundation with OSS. Most of a company's software was never something to build; it was something to buy. Microsoft 365, Copilot, WordPress, the vendor packages under the core systems. What replaces them already runs all over the world.

• The machine: one Debian box (2-02) — the place you hand to the AI
• The foundation: SQLite, PostgreSQL, DuckDB, Polars (2-03)
• Authentication: PocketBase, in place of Entra ID (2-05)
• Sharing and version control: Forgejo, in place of GitHub (2-06)
• Documents: things you read held as AsciiDoc text, tables you work in in a grid, pages you print from templates (2-07)
• Mail: Stalwart (2-08); meetings: Jitsi (2-09)
• Web: baked HTML on your own machine with Caddy, or on Cloudflare Pages, in place of WordPress (2-11)
• Core logic: exposed as an API with FastAPI, in place of the vendor package (2-12)
• AI: a local LLM plus RAG, in place of Copilot (2-16)

None of this means every piece is finished today. Documents are not a tool to stand up but a way of holding content, and the grid can stay Excel (2-07). For publishing the web, your own machine keeps everything on your side, while Cloudflare Pages means borrowing only the window. Decide separately what to borrow and what to hold — that is what laying the foundation consists of.

First, untie the generic from the vendor. Then put the foundation — identity, data, documents, mail, web, AI — on your own side. On top of that, write only the logic that is genuinely your company's own, together with AI. Writing code comes in the later half, after the foundation is laid. And the amount to write shrinks to the company-specific part.

This foundation-building is the Independence part, Part 2 of this series. One OSS stack at a time, you become independent of Microsoft 365, Copilot, WordPress, the core systems, and GitHub.

What AI cannot do, the SIer cannot do either

Turn all of this over and one strong consequence falls out. The generic can be built on OSS, and the specific can be built with AI by yourself. If so, where is the reason to commission an SIer for the whole thing?

The old reason to hire an SIer was "we cannot build it ourselves." But in the AI-native world, the AI the SIer uses and the AI the customer uses are the same AI. The top-tier model is sold to anyone at the same price (1-01). There is no specially stronger AI that only an SIer can reach. So what AI cannot do, the SIer cannot do either. A problem AI cannot solve gets the people inside the SIer stuck the same way, because they use the same AI.

The SIer's real advantage remains in experience and judgment in the areas AI cannot reach. Genuinely new technology, specialized regulation, cross-organizational negotiation, the pitfalls you only learn from experience. But that is a small slice. Taking it in as advice is enough. It is the same as not handing routine business wholesale to a lawyer or a tax accountant (3-06). The multi-year SIer commission is no longer needed.

Summary

The customer builds the generic on OSS and the specific with AI.

• The generic runs on OSS — do not write it, do not buy it. It is the most economical path, and it is hardened as a defense too
• The personal runs on OSS plus AI — add yourself on top of the foundation, in everyday words
• An organization starts from the foundation — machine, data, identity, code, documents, mail, meetings, web, API, AI, all on your own side
• Only the specific gets written with AI — writing code comes in the later half, once the foundation is laid
• What AI cannot do, the SIer cannot do either — because they are using the same AI

That ends the Introduction part. AI became the strongest SIer (1-01); the unit of maintenance moved to context (1-02); the roles of design and code moved to the builder (1-03, 1-04); and the customer now stands on the building side (1-05). What remains is only the question of what to actually stand up, and how.

The next chapter opens the Independence part. 2-01: Becoming Independent from Microsoft and Google — The Whole Map lays out the whole map and the replacement table, and from there, one chapter at a time, the machine, the data foundation, identity, code, documents, mail, meetings, web, APIs, diagrams, electronics, and AI all move to your own side.

Related articles

• 1-01: AI Solves the World's Hardest Coding Problems
• 1-03: AI Now Does the Software Engineer's Work
• 1-04: The Builder Role
• 2-01: Becoming Independent from Microsoft and Google — The Whole Map
• 3-05: The Lock-In Problem
• 3-06: Companies Hire Builders
• Structural analysis 08: Subtracting the enterprise-IT tax
• Structural analysis 12: AI and the sole proprietor