--- title: "The barbarians are already inside the gate" url: "https://barbairians.com/blog/barbarians-at-the-gate-of-enterprise-software" description: "The barbarians of 1988 borrowed money and bought the company. The ones at the gate now build systems that make the company they are storming irrelevant." --- # The barbarians are already inside the gate August 27, 2026·3 min read·BarbAIrianS In 1989, Bryan Burrough and John Helyar gave the corporate establishment a name for the thing coming through its gates. _Barbarians at the Gate_ told the story of the RJR Nabisco buyout: financiers with borrowed money storming a company that had grown comfortable, and winning. The phrase stuck because everyone in a boardroom recognised the feeling. Thirty-odd years later the gates belong to somebody else — the hyperscalers, and the ERP vendors whose software runs the companies that buy from them. There are barbarians outside those gates too, and they want something different. ## What the first barbarians never touched The 1988 raiders changed who owned RJR Nabisco. They did not change how it worked. The same processes ran, on the same systems, executed by the same people; only the capital structure and the letterhead were different. That is the limit of a financial takeover — it is a transfer of ownership, not a change in how the market works. The barbarians at the gate now are not buying anything. They are small owner-operated companies running systems the incumbents cannot sell them, and the threat they pose is not to a share price. It is to the assumption that running a serious business requires renting eleven products from people who have never seen your warehouse. ## What actually got outsourced Not the servers. Nobody sensible wants to run their own mail server. What got outsourced is subtler and much more expensive: the ability to change your own business. Every process that lives inside a vendor’s product moves at that vendor’s speed. You want a field added, a rule changed, a report that joins two systems that were never meant to meet — and the answer is a roadmap, a professional services quote, or a “that’s not how the platform works.” Multiply that across eleven products and the compound effect is a company that can only do what its software already anticipated. That is not a tooling problem. It is a loss of capability, and it looks exactly like hiring mercenaries because fighting is beneath you. ## Why small teams are suddenly dangerous For thirty years, scale was the answer to coordination cost. A large company could afford the specialists, the integrations and the process overhead that a small one could not, and that advantage compounded. AI-native systems attack precisely that overhead. When an agent can read the contract, reconcile the invoice, draft the response and flag the exception, the ratio between what a team can decide and what it can execute changes. The specialist layer that justified the size of the organisation gets thinner, and the small owner-operated company — the one that was always faster at deciding — stops being slower at executing. That is the whole barbarian thesis. Not that big companies are stupid, but that their main structural advantage is being eroded by tooling that costs a fraction of what their licences do. ## Owned data compounds, rented data expires Here is the asymmetry that decides the next decade. A company that keeps its operational history — every order, every exception, every decision and its outcome — inside systems it controls can put that history to work: fine tune on it, mine it, replay it, and hand it to whatever model exists in three years. A company whose history lives in eleven SaaS products has eleven partial exports in eleven schemas, most of them missing the very context that made the data valuable. When the contract ends, so does the asset. You were not building a data advantage. You were renting one, and paying to improve someone else’s. ## What this is not It is not a call to self-host everything. Running your own identity provider to prove a point is how a five-person company spends its year not shipping. The argument is narrower and more useful: **know which layers decide your margin and your survival, and own those.** Hardware where jurisdiction matters. Models where the capability is core. Data always. Orchestration if the process is your actual product. Rent the rest happily, on short contracts, with the exit tested. ## Why now and not five years ago Three things changed at roughly the same time. Open-weight models became good enough for real operational work rather than demos. The hardware to run them stopped being exotic. And European regulation started treating portability and data access as rights rather than as features vendors may generously offer. None of those alone would matter. Together they mean the sovereign option is no longer an ideological luxury with a capability penalty attached — which is what it was, honestly, until quite recently. The barbarians of 1988 needed an investment bank. This lot needs a weights file, a decent server and the nerve to cancel a subscription. - sovereignty - AI-native - competition - strategy ## Keep reading - [Sovereign AI is procurement, not patriotism](https://barbairians.com/blog/sovereign-ai-is-procurement-not-patriotism) - [Open weights are the means of production](https://barbairians.com/blog/open-weights-are-the-means-of-production)