What does it take to make an AI agent work in real-world software development? At the latest Let’s Talk session, Viet Anh Vu from Umbraco shared his practical approach to working with AI, moving the conversation beyond models to the systems and human decisions that make them useful.
Agent = Model + Harness
Held on 16 July 2026 at Scandinavian Software Park, the session welcomed 34 registered participants from across the SSP portfolio companies.
Under the topic “Agent = Model + Harness: How I Actually Work,” Viet Anh shared the workflow he uses when coding with AI, covering prompting, context management, tools, and reviewing outputs from multiple perspectives.
The discussion highlighted a broader view of AI agents. A capable model is only one part of the system. The surrounding “harness”, including prompts, tools, workflows, context, and evaluation, plays an equally important role in turning AI capabilities into solutions for real development tasks.
Viet Anh also shared how approaches such as Superpowers and BMAD can be incorporated into a more unified working process.
Keeping Humans in the Driver’s Seat
Another key theme was the balance between AI autonomy and human direction.
Rather than giving AI full autonomy, Viet Anh shared why he keeps his daily development workflow human steered. AI can support different stages of the process, while engineers remain responsible for reviewing results, bringing different perspectives together, and making the final decisions.
The session also looked beyond individual workflows to consider how these practices can scale from solo development to team collaboration, particularly as tasks become more complex.
