Creating a market watch agent to spot commercial opportunities early

Context and objectives

A leading demolition and deconstruction company wanted a reliable way to identify commercial opportunities in the site-reconversion market: large companies whose buildings could be demolished to make way for redevelopment. The team had already written a broad, all-in-one prompt and was running it manually across several AI tools.

This approach ran into clear limits:

• The prompt tried to scan everything at once, from research to scoring to formatting

• Results varied significantly from one run to the next

• Success was difficult to measure

• The process had no clear finish line or defined output

The objective was to move the team from manual, scattered prompting to a more structured method, while giving them enough hands-on grounding to design, build, and evaluate an AI agent for their own market-watch use case.

Approach

1. Prompt engineering workshop

The first workshop focused specifically on prompt engineering for the client's commercial-watch use case. Working hands-on, the team:

• Reworked their original, all-in-one prompt

• Split it into smaller, targeted sub-prompts, each with a single clear task

• Compared outputs across different AI providers through practical exercises

This gave the team applied experience in why breaking a complex task into distinct steps produces steadier and more measurable results than a single, monolithic prompt.

2. Agentic platforms workshop

The second workshop broadened the scope to the technical foundations needed for a more automated solution. It covered:

• A shared vocabulary around language models, tools, and agents

• A comparison of code-first and low-code development frameworks

• A comparison of cloud versus on-premise hosting, including cost considerations

• A concrete outline of a potential automated watch pipeline

The workshop closed with three live demonstrations of the same demolition-watch task, each built with a different technical approach, giving the team a side-by-side view of the trade-offs involved.

3. Building the market watch agent

With both workshops delivering clear results, the engagement moved into a dedicated development phase to build a working market watch agent. This phase put the concepts from the workshops into practice, translating the client's ambition to scan broadly for opportunities into a design that stayed as structured and traceable as possible.

The result was not only a functioning agent but also a reusable blueprint the team can apply to other agentic use cases going forward.

Results

The engagement delivered two structured workshops and a working market watch agent, giving the client both a hands-on foundation in prompt engineering and agentic platforms, and a concrete, reusable outcome.

Qualitative outcomes:

• The team moved from an unpredictable, single-prompt approach to a structured method with measurable steps

• They gained practical experience comparing outputs across multiple AI providers

• They developed enough technical grounding to evaluate agent-based and hosting options with confidence

Business impact

The final phase delivered a working automated agent, built with a low-code framework, directly applying what the team learned during the workshops. Beyond the agent itself, the client walked away with a reusable blueprint for future agentic projects, reflecting the core value of the phased approach: each stage built the team's confidence and knowledge, enabling an informed decision to invest further in automation.