Can non-technical staff build AI agents? What three days should produce

Who makes a good builder?

The best builders are rarely the most technical people in the room. They are the people who know a process step by step: where the inputs come from, which checks matter, where things usually go wrong and who signs at the end. An agent is a written-down version of that knowledge, with a model doing the reading and drafting.

The research points the same way. In a study of 5,172 customer-support agents, an AI assistant raised issues resolved per hour by 15%, and newer, less experienced staff gained most (Brynjolfsson, Li and Raymond, Quarterly Journal of Economics, 2025). The tool narrowed the gap between novice and expert. Building is similar: the platform supplies the technique, so the scarce input is process knowledge and judgment.

Starting skill is low almost everywhere. In one enterprise benchmark, 13% of employees were assessed as accomplished in agentic AI before training (Workera, 2026 AI Skills Enterprise Benchmark, 88,753 assessments; Vendor data). That is not a reason to wait. It is a reason to select builders for what they know about the work, not for what they already know about AI.

Good candidates usually share three traits: they own or run a repeated process, they can describe a good and a bad output, and their manager will give them time to finish.

What changes on each day?

In Cader Group's Agentic Build program, the Accelerated format runs over three days. Each day has its own output.

• Day 1: the use case, on paper. Builders map one process: inputs, steps, the output, and the point where a person approves. They decide what the agent may do and what it must hand back. If the agent is really a question-answering tool, this is where it shows; our note on AI agents and chatbots explains the difference.
• Day 2: the build. Builders assemble the agent on a no-code platform, connect it to approved documents and sample data, and test it on real past cases. Most of the day goes on instructions, checks and the cases the agent gets wrong.
• Day 3: tests, controls and the demo. Builders test edge cases, set what the agent must refuse or escalate, demo it on a real past case to the people who would use it, and write the handover plan.

No programming is required at any point. What is required is the discipline to test on real cases rather than the easy ones.

What are the capstone and the handover plan?

The capstone is a demo of the agent running on a real past case, in front of the people who would own and use it. A demo on invented inputs proves little; a demo on last quarter's work shows where the agent helps and where it stops.

The handover plan is what turns a demo into a decision. It names:

• The owner who will run the agent after the program.
• The approval route: who signs each output before it takes effect.
• The data and tools the agent needs, and who has approved them.
• The next step: a live pilot, further testing, or stop.

That plan matters because building is not the hard part of agentic AI; keeping it alive is. Gartner forecasts that over 40% of agentic AI projects will be cancelled by the end of 2027, due to escalating costs, unclear business value or inadequate risk controls (Gartner, June 2025; forecast). A named owner, a stated value and an approval route answer all three before the project starts. Agents worth taking further can move into our AI Pilot Program, where they run live against a baseline for six to eight weeks.

What must the organisation provide?

Most programs that disappoint fail on inputs, not teaching. Four things should be settled before day 1:

1. A use case. A real process the builder is free to act on, frequent enough to matter and with a clear output. The final choice is made on day 1, and the facilitator redirects it early if it cannot be built in the time.
2. An owner. A named person who will receive the handover plan and decide on the next step.
3. Data access. Sample documents and past cases that mirror real records, cleared for use in the program.
4. Approved tools. A no-code platform and model the organisation has approved. Where the approved stack cannot support the build, a sandbox can be used, and the handover plan states what must move onto the organisation's own stack before a pilot.

If any of the four is missing, the program becomes a demonstration of a platform rather than a build.

Accelerated or Complete: which format?

Agentic Build runs as Accelerated (3 days) or Complete (8 days). Accelerated is narrower, not compressed: it takes one use case and one agent pattern from design to demo. Complete covers more ground at the same depth, including agents that hand work to each other, connections to more systems, and stronger controls and monitoring, so builders leave able to design the next agent as well as this one.

Choose Accelerated when you have one clear use case and want a decision on it quickly. Choose Complete when you are building a small group of internal builders who will support several teams.

Measure where your own team stands: take the free Readiness Snapshot →