Will be speaking at a panel at Mirren live in May 13 and 14

I spoke on a panel at Mirren Live in May, on what an agency should build for itself when the large AI companies ship new features every week and are absorbing whole categories of startup as they go. Rebuilding what they already hand you is not the answer. At Tombras we built twelve agents grounded in our own proprietary data instead: brand health measures like T-Score and sentiment, creative tools including a brief generator and a big ideas generator, audience intelligence, and a retrieval layer over our internal documents.

I finished by sharing some of what building it actually taught us. A few of those:

• Constantly reassess build versus buy. The ground moves every quarter.
• It is about outputs and outcomes, not about the agents themselves.
• Intent has to travel from stakeholder to engineering to agent without being lost along the way.
• Subtle architecture choices have outsized impact, such as whether a thing is a sub-agent or a tool.
• Rich data, public and proprietary, is the foundation of any decent output.
• Human oversight and iteration are not negotiable at the speed and volume these systems run at. AI accelerates ideation; people curate for tone, accuracy and taste.
• Play with everything, but put only a few things into production. Siloed AI does not transform an organisation.