AI that does something useful.

We do not solve every problem with AI.

Every prospect we speak to about AI starts with "can you build us an AI thing?". Our first question is always "does this actually need AI?". A well-written regex beats a poorly-used LLM. A scheduled job beats a chatty agent. The work we ship with AI uses AI because it is the right answer, not because it is the premium answer.

Things AI is good for:

• Unstructured-to-structured conversion (documents, forms, emails, transcripts)
• Natural-language interfaces to existing systems
• Bulk operations that need judgement
• Classification and routing (triage, tagging, summarisation)

Things AI is not yet good for:

• Decisions where being wrong is expensive and unreviewable
• Tasks that depend on knowledge you have not given the model
• Anything that looks better as a lookup table or a deterministic rule

If your site runs on Umbraco, AI is closer than you think.

Eddi gives your editors a conversational interface to your content. Works on v13 LTS and v17. Installs via NuGet. Plugs into Anthropic Claude, OpenAI, or any Microsoft.Extensions.AI-compatible provider. BYOK - your key, your cost control.

Umbraco 17 also ships Umbraco.AI natively, which handles the single-field AI operations that editors use on individual pages. We wrote a comparison: Eddi vs Umbraco.AI. The short version: the tools don't fight, and most agencies running both get the best of both.

AI beyond Umbraco.

Not every client runs on Umbraco. Our AI Integration service applies the same patterns to any .NET application, and to most non-.NET stacks we audit into. Typical work:

• Content AI assistants custom to your CMS or application
• Document pipelines with intake, extraction, classification and delivery
• Custom MCP servers that give Claude (or any MCP-compatible model) structured access to your systems
• AI-driven workflows running on a schedule or trigger, with human-in-the-loop where it matters

Engagements run from a single £500 discovery day through to £20,000 scoped builds. Model costs sit on your account directly - we never mark up API usage.

We do not just integrate AI. We build with it.

Claude Code has been our default development tool since 2025. It writes the boilerplate - data models, CSV parsers, webhook handlers, unit tests - while we write the architecture and make the decisions that matter. It is the reason Sumlo shipped in three days and Patch shipped in three weeks.

This is not a service we sell separately. It is baked into Rapid Development and every other project. If you hire us, this is how we work.