Fieldwork

Built: An AI agent that acted as a QA sparring partner before we had a dedicated QA role, first keeping work moving through QA, later reviewing backlog items and adding acceptance criteria.

Learned: An agent brought QA discipline into the flow of work long before we could afford a QA role. Not as good as a senior QA, but much better than what we had before.

Built: A tool that sends open-ended questions to hundreds of users and uses AI to find themes, insights and outliers in their free-text answers, removing the old choice between deep interviews and narrow surveys.

Learned: The cost of learning from customers can drop, not just the cost of building. The work moves to asking open questions and judging what comes back. questions.

Built: A workflow that collects real questions managers post on Reddit, and an AI model that scores how well each one fits the product's domain, with a written reason for each score.

Learned: You can learn from users before you have many. Real questions from your target audience give a steady stream of problems to test the product against.

Built: A development process where discovery, design and build each hand over to the next stage through a formal document, used to build a working web application end to end, to test whether those handoffs carry enough context for different specialists.

Learned: Formal handoffs can keep each specialist's questions inside their own role. Making sure people actually review them is a separate problem.

Built: A structured set of product knowledge files for Work2gether AI, created by having an AI agent study the existing material and interview me until everything agents need to work on the product was written down.

Learned: Even a small product holds far more unwritten knowledge than anyone realizes. An agent can draw it out quickly, but someone still has to own keeping it true.

Built: Brought AI-assisted building to an existing demo version of where the product was heading, used with prospects, leadership and the board, but too costly to maintain alongside the real product.

Learned: Because the demo never had to hold up in production, AI made it cheap enough to keep. Something with proven value to the business became affordable at our size.

Built: Automated sign-up, email validation, secure trial provisioning through the product's API, two weeks of onboarding emails and ongoing nurture emails, so the business runs without me in the loop.

Learned: When building is cheap and operator time is the scarce resource, automating operations belongs in the first version, not after product-market fit.

Built: A secure portal that lets people use AI agents through one chat interface, signing in with their Microsoft 365 account, with access controlled per organization, user and agent.

Learned: The agent was the fast part. Making it secure and usable for others was the real work, and a shared layer meant doing that once instead of for every agent.