What Happens When AI Agents Become Colleagues?
Jesse Vincent describes running a small company alongside persistent AI colleagues with roles, responsibilities and shared processes. In conversation with Shareen Pathak at BRXND NYC 2026, he explains why the experience draws heavily on management: making intent clear, setting expectations, reviewing outcomes and deciding what is worth shipping.
Key takeaways
- Explain the intent and desired outcome behind a task, alongside the process you expect an agent to follow.
- Give persistent agents clear roles and feedback, and evaluate the work they actually deliver.
- Keep taste and judgment involved when deciding which experiments should become shipped products.
- Use failures to improve the instructions and revisit capabilities as the tools change.
In this session
00:00 Working at the edge of AI with curiosity
Pathak and Vincent open with curiosity and the difficulty of keeping pace with a rapidly changing field.
01:01 A year of building software without writing code
Vincent describes a year of building through agents and compares the work to managing engineers through text communication.
05:49 The human role: taste, judgment and intent
He identifies taste, judgment and intent as the human contribution that gives the work a particular point of view.
06:36 Why capable models still need structure
The discussion turns to why an organization still needs to provide its own processes and expectations as models improve.
07:52 Agents versus persistent AI colleagues
Vincent distinguishes task-oriented agents from the persistent AI colleagues his team is experimenting with.
10:48 How AI colleagues collaborate
He shares examples of those colleagues coordinating work, noticing problems and interacting with human team members.
13:45 Turning management experience into agent skills
Vincent explains agent skills as written processes with the reasoning and intent behind them.
14:50 Empowered teams and deciding what to ship
The speakers consider what small empowered teams can do and distinguish permission to experiment from the decision to ship.
18:05 Practical advice to try tomorrow
Vincent recommends trying the tools again, asking why an attempt failed and using that feedback to become clearer about what you want.
Recorded at BRXND NYC on November 5, 2026, at The TimesCenter in New York.
Written by
.png&w=3840&q=75)
Alephic Writing
Ideas for building the AI-native enterprise.
New essays on AI, marketing, software, and the organizations being rebuilt around them.
