"I Can't Do That, Dave" – No Agent Yet
3 days ago
- #AI Agents
- #Software Engineering
- #Collaboration
- The industry is focusing on building agents that quickly say 'yes', but sometimes the correct answer should be 'no'.
- Isolation in software engineering leads to flawed systems, a lesson repeatedly forgotten over fifty years.
- Meaning and quality outcomes emerge from conversations, not isolation, as highlighted by various experts across different fields.
- Current AI prompting interfaces are considered poor and should be phased out for better interaction models.
- Agents lack continuity between sessions, which is crucial for meaningful participation and problem-solving.
- The real challenge is not giving agents memory but shaping the ground between sessions for persistent identity and stakes.
- Historical lessons from software engineering emphasize that isolation degrades system quality and collaboration is key.
- The industry is reverting to siloed workflows with AI, reminiscent of Waterfall, under the guise of 'AI-native development'.
- Persistent identity and participation, not just memory, are essential for agents to have a stake and push back effectively.
- The proposed solution involves integrating agent identity with code persistence systems like git, using cryptographic methods for trust.