The evidence on internal builds is not close, and the wins that are available without building are larger than most firms expect.
Start a conversation with the AI Adoption Concierge, already scoped to automation & building. Choose the question closest to yours, or describe your situation directly.
There is a persistent assumption that getting serious about AI means building something. The evidence points the other way: research across enterprise deployments has found internally built systems succeeded roughly a third as often as buying from specialists, and the failures cluster around integration, data quality and ownership rather than model capability. Meanwhile the automation that reliably pays for itself in a law firm is unglamorous and largely pre-dates AI — matter opening, e-signature chasing, deadline synchronisation, review requests. The genuinely new development is connection: standards emerging through 2025 and 2026 mean a firm's document system and research provider can now be reached by whatever AI it uses, without a bespoke integration project. That is configuration, not construction, and it is where the leverage is.
The automation worth doing first, connecting what you already own, and the decision about whether to build at all.
The unglamorous wins that pay back in weeks, most of which need no AI at all.
investigateThe 2026 development that matters most, and why "integrates with" tells you almost nothing.
investigateA third option most firms miss, and the honest failure rate on the one they reach for.
investigateHow the Institute approaches this.
Matter opening, almost always. It is high-frequency, touches several systems, has a clear beginning and end, and every step except the conflicts decision is mechanical — intake data captured, matter and contact created, folder structure built, engagement letter generated and sent for signature, standard tasks and deadlines created, responsible attorney notified. The conflicts check stays a human gate in the middle and the workflow pauses there. Firms that automate this typically see it pay back within weeks, and it is the one that teaches the firm how automation behaves before anything riskier is attempted.
For most definitions of "build," yes. Research across enterprise AI deployments found internal builds succeeding at roughly a third the rate of purchases from specialist vendors, and the failure modes are consistently organisational rather than technical — nobody owns it, the data was worse than assumed, the person who wrote it left. What that does not mean is that firms should do nothing. The productive middle ground is configuration: shaping a vendor product or an open framework with prose, prompts, playbooks and connectors rather than code. That is where firms with no engineering capacity have produced real results.
It is a standard for connecting AI applications to the systems that hold data, open-sourced in late 2024 and donated to a neutral foundation in December 2025. It matters because two dominant law-firm document management systems shipped support for it in 2026, along with research providers and a body of court data. The practical consequence is that a firm can point an AI tool at its own governed document repository without commissioning an integration — and that the connection inherits the existing permissions rather than bypassing them. It also means an agent can be given tool access, which is a security question covered elsewhere.
For most of what is worth doing, no. Workflow automation between systems is a configuration exercise a capable operations or knowledge person can run, and the platforms are priced in tens of dollars a month rather than thousands. Prompt and template libraries need editorial discipline rather than engineering. Where a developer genuinely becomes necessary is building retrieval over the firm's own document corpus with permissions enforced correctly, and running agent frameworks — both of which are things most firms should be buying rather than building anyway.
Describe what you are trying to change. The Institute will help you work out which it is.