From Better Prompts to Trusted AI Agents: Clare Kitching’s Seven-Step Framework

In a recent LinkedIn post, Clare Kitching discusses how professionals can move from asking AI questions to trusting it with complete, repeatable work tasks. This article reports on Kitching’s framework and examines the gradual increase in responsibility she describes; it is coverage of her post, not an article written by Kitching.

“It’s a big jump from asking AI a question to trusting an agent with a piece of your work.”

A progression from chat to context

Kitching frames AI adoption as a series of practical steps, using a Monday business update as her example. At the starting point, a user provides recent figures and asks a chat tool for a first draft. The next step is to preserve background information in a project, reducing the need to explain the team and its progress each time.

According to Kitching, those early improvements can save effort, but they leave users responsible for finding the right spreadsheet and rewriting commentary. Her framework suggests that greater delegation depends on connecting AI to relevant information and defining the task clearly.

Giving AI the data and the task

At step three, Kitching proposes connecting AI to the source data so it can retrieve current figures and distinguish expected results from actual ones. Step four expands the assignment: instead of asking for a draft, the user asks for a usable meeting file organized around specific business needs.

“Give it the whole task.”

That shift, as Kitching presents it, changes the user’s role from repeatedly assembling material to reviewing an output. Her example asks an agent to group customers by account manager, show their usual order value and flag ten accounts for discussion. The resulting file still requires human review, but it is designed to support the meeting directly.

Making a task repeatable

Turning tacit habits into instructions

Kitching identifies step five as the key advance: translating a person’s method into a reusable skill. She points to actions such as comparing actual results with budget, highlighting the largest variance and writing commentary in a familiar style. These routines may feel automatic to the person doing the work, she argues, but an AI system needs them made explicit.

“For me, step 5 is the big unlock and once you start developing skills, you can’t go back.”

Once the process is reliable, Kitching’s sequence moves to scheduling the update for Monday morning. The seventh step extends the agent’s role further: checking notes for explanations behind variances and following up with colleagues when those notes do not account for a gap.

Progress through a familiar task

Kitching’s post presents agent adoption as an incremental change in working habits rather than a single leap. The sequence moves from drafting, to using persistent context, to accessing source data, completing a defined task, applying a learned method, and eventually running on a schedule. Each stage adds responsibility while keeping the user involved in shaping and reviewing the work.

Her closing recommendation is to try moving one task up the steps. For teams exploring AI agents, the framework offers a concrete place to begin: choose a recurring piece of work, identify the method behind it and make that method clear enough to repeat.