Can your agents actually ship? was part of my first day at SF Tech Week. Its focus on delivery beyond code generation connected with the questions I was beginning to ask about scaling agents.
The program covered pipelines, CI, review and security gates, context, validation and governance.
Look at the delivery path
The program framed shipping as more than generating code. The question I want to carry forward is what happens between an agent’s output and a change being accepted into a real product.
Ask how a result is checked
Review, validation and security gates were part of the advertised discussion. For me, those topics connect with learning about evals across the week: a result needs a way to be examined before a team relies on it.
Keep context and governance in view
Context and governance were also program themes. My follow up questions are what information an agent has, which decisions it can make and who is responsible for accepting the work.
My takeaway
When evaluating an agent workflow, ask about the path to delivery as well as the generated output. The program gave me a useful set of questions to explore further.
Thank you to the hosts and organizers for a conversation about what it takes for agents to ship.
Questions from this recap
What was the event’s focus?
The advertised program focused on delivery beyond code generation, including pipelines, CI, review, security gates, context, validation and governance.
What should I ask about an agent’s path to delivery?
What happens after output is generated, how it is reviewed and validated, which security gates apply and who accepts the finished work?
Explore further
What must happen before your team accepts an agent’s output as a finished change?