What does "production AI" mean here?

It means the AI is part of a deployed integration that operates on real data, has measurable outcomes, has cost and safety controls in place, and is in the on-call rotation. Not a sandbox, not a slide deck, not a pilot that never crosses the line.

Yes. Function-calling and tool-use against operational APIs is part of most engagements. We design the tool surface, the safety boundaries and the audit trail alongside the agent itself.

How do you handle PII and data privacy?

We design the data surface up front. PII is masked or redacted before model calls where appropriate, EU-hosted models are available on a per-turn basis for engagements that require it, and every model call is logged against the operator and the cost centre.

Which models do you work with?

Frontier models from the major providers (Claude, GPT, Gemini), plus open-weights where the use case justifies the operational overhead. Model selection is made against the outcome, the latency budget and the cost ceiling, not as the headline of the engagement.

Service

AI enablement

AI enablement is integration work, with AI in the loop. We help businesses ship production AI against the systems already running their operation, not run another pilot.

Talk to us about ai enablement See all services

The practice

How we run ai enablement.

The AI enablement category gets a lot of pilots and far fewer production deployments. Our practice is the production end of that distribution. We help businesses add AI to systems they already run: agents that operate on live operational data, function calling against internal APIs, retrieval pipelines over commerce and customer records, embeddings and vector search where they earn their place in the architecture rather than being added because the category demands it.

Concrete examples of what we ship: customer service automations that read order context from the ERP, summarise the issue and propose an action; product description generation that pulls structured attributes from the PIM and writes channel-specific copy with the brand voice respected; agent workflows that handle returns triage against operational rules; retrieval-augmented assistants that answer internal questions over policy and process documents; classification and routing on inbound emails, tickets and supplier communications. The pattern is the same in each case. AI is one node in an integration that touches systems already running the business.

The discipline we bring is the discipline we bring to any integration. Function calls have rate limits and cost ceilings. Retrieval has freshness and lineage. Models change and prompts drift. PII has consent boundaries that span jurisdictions. Agents need audit trails for the operator who ran them and the cost centre that paid for them. These are integration concerns dressed in different vocabulary, and the engineering rigour that makes a Shopify-to-NetSuite flow reliable is the same rigour an AI agent needs to be production-grade. We bring both halves to the table.

We scope AI engagements by the operational outcome rather than the model choice. The first conversation is about what the business wants to be different in three months: faster ticket resolution, fewer manual product description hours, better self-service answers on the storefront, agent-assisted scoping for the next integration project. From there we work back to the integration design, the model selection (frontier model, open-weights, hosted region), the data preparation, the safety surface and the on-call shape. The model picks itself once the outcome and the constraints are agreed.

PatchBuddy, our own AIiPaaS product at patchbuddy.ai, is the proof point: an AI agent in production on a Patchworks tenant since January 2026, used by our delivery team and licensed to other Patchworks agencies and customers. The engineering disciplines that ship PatchBuddy are the same ones we bring to client AI engagements.

Common platforms

Platforms that come up most often.

Questions

Common questions.

01 What does "production AI" mean here?

It means the AI is part of a deployed integration that operates on real data, has measurable outcomes, has cost and safety controls in place, and is in the on-call rotation. Not a sandbox, not a slide deck, not a pilot that never crosses the line.
02 Do you build agents?

Yes. Function-calling and tool-use against operational APIs is part of most engagements. We design the tool surface, the safety boundaries and the audit trail alongside the agent itself.
03 How do you handle PII and data privacy?

We design the data surface up front. PII is masked or redacted before model calls where appropriate, EU-hosted models are available on a per-turn basis for engagements that require it, and every model call is logged against the operator and the cost centre.
04 Which models do you work with?

Frontier models from the major providers (Claude, GPT, Gemini), plus open-weights where the use case justifies the operational overhead. Model selection is made against the outcome, the latency budget and the cost ceiling, not as the headline of the engagement.

Get in touch

Tell us what you’re trying to connect.

And what’s in the way. We will tell you whether we are the right people to do it. Drop us a line below, or open the chat in the corner of the screen.

Direct: [email protected]

AI enablement

Tell us what you’re trying to connect.

Thanks. We’ll be in touch.