AI
AI Is Changing How We Learn at Work
- AI “shortcuts” disrupt natural pathways of mastery (practice, frustration, honing the craft) are skipped. “…but accelerated learning is not the same as development. Acceleration increases output; development transforms identity. The two are not interchangeable.”
- Just like COVID-19 accelerated the adoption of digital-collaboration platforms led to more meetings and intense workloads, AI expands the volume of the content (presentations, drafts, reports). “We’re generating more but thinking less.” Noise is drowning out calm, reflection, and the space to think deeply.
- Empathy grows through practice, so what happens when AI handles the tough stuff (e.g., difficult conversations, supporting others)? Convenience removes the conditions where empathy, judgment, and relational capability are formed.
- “‘If the system always knows the next step,’ one executive asked me recently, ‘when do my people learn to choose for themselves?'”
Leadership in AI Assisted Engineering – Justin Reock, DX
- A recent METR report showed differences in perceived vs actual performance. (There may be some flaws in this report.)
- DORA put out a report based on industry averages. There are modest, positive changes. DX (productivity measurement company) reviewed its own data and found the same thing. What’s interesting is that the variability in the data was high where large gains are erased by large losses.
- ROI is not evenly distributed in terms of KPIs, adoption, top-down mandates (which impact psychological safety), lack of education and enablement, expectations of users to be proficient, difficulty measuring the impact.
- Integrate across the SDLC. Code is not the bottleneck.
- Unblock usage by seeking ways to limit barriers to adoption (e.g., on-prem models, local training).
- Have open discussions about metrics and why they’re important. Showcase wins.
- Reduce the fear of AI (it augments, not replaces).
- Validate AI output using HITL and leverage existing practices such as TDD and linting.
- Tie career success to leveraging AI.
- Google’s Project Aristotle found the psychological safety is associated with high-performing teams. Explain why AI is part of the solution.
- Increase PR throughput but not by creating slop that leads to tech debt. Look at change failure rate, perception of quality, change confidence, maintainability, % of time allocated to bugs.
- API telemetry metrics — good for measuring impact on developer output, not good for understanding how the tools are used (e.g., accept AI changes then edit afterword).
- Developer experience is a systems problem, not a worker problem.
- The DX survey looks into metrics around utilization, impact, and cost. These dimensions are also a maturity measure.
- It’s helpful to have a group that takes ownership of system prompts and incorporates feedback from teams using them.
- Be mindful of non-determinism and model temperature (where 0.1 to 0.9 makes it more deterministic). Where do you want more or less creativity?
- Provide education and an adequate time to learn.
- Find actual bottlenecks in the system and find ways to leverage AI.
Information
Nexus: A Brief History of Information Networks from the Stone Age to AI
This was a thought provoking book about how information moves in groups of humans. There’s history, philosophy, ethics, and political science. Something different from previous forms — stories, holy texts, the printing press, the industrial revolution — is that AI is the first tool that’s capable of making decisions and generating ideas by itself. Every tool has the possibility to positively impact humanity in profound ways, or it could threaten our very existence.