"We want to explore agentic AI" is not, on its own, a scope. It can mean three very different projects, with very different budgets, timelines and outcomes — and the single most common way agentic AI budget gets wasted is starting the wrong one.
If you don't yet know whether the use case is viable: prototype
A rapid prototype exists to answer one question — does this actually work on our data, in our workflow — as cheaply and quickly as possible. Two to four weeks, a small focused team, a working demo against real (not synthetic) data, and an explicit go/no-go decision at the end. If the honest answer might be 'no,' this is where you start. Skipping straight to a full build because a use case sounds obviously right is how six-figure budgets get spent proving something that could have been tested for a fraction of the cost.
If the use case is validated and needs to run in production: bespoke development
Once a prototype has proven the case, the job changes: now it needs to survive contact with real users, real load, real audit requirements, and real regulatory scrutiny. That's a governed build — protocol-native architecture, guardrails, an embedded team working alongside yours, and knowledge transfer built into the contract rather than left for the end. This is the phase where 'it worked in the demo' has to become 'we can defend this decision to a regulator.'
If you're building your own AI-native product: product development
Different problem again. Here the client isn't automating an internal workflow — they're building something they intend to sell. That needs full-lifecycle product work: experience design, engineering, a launch plan, and a support model, not just a working backend. The governance foundation is the same, but the deliverable is a product, not a workflow.
A quick self-check
- Could this use case plausibly fail once you actually try it? → Prototype first.
- Has it already been proven, and does it now need to run reliably at scale? → Bespoke development.
- Are you building something you intend to sell to your own customers? → Product development.
The good news: none of this is wasted if the sequencing is right. A well-run prototype's architecture, data connections and governance decisions carry directly into the bespoke build that follows — nothing gets thrown away, it just gets extended.
Not sure where your use case fits?
Take the five-minute Agentic Readiness Assessment for a tailored starting point.