Leadership
Leaders Need to Grapple with Employee Ambivalence on AI
- “…it’s moving faster, touching every level of organizations at once, and forcing decisions before anyone has a real playbook.”
- “As a leader, your job isn’t to project a certainty or positivity you don’t have. It’s to acknowledge the tension, create a safe sandbox, and reinvest the gains.”
- AI without experience a human brings is generic.
- Don’t force people to be purely positive. Address the elephant in the room.
- Trade anxiety-inducing urgency for curiosity-based experimentation.
- “…help them ‘experience an instant productivity boost’ with AI and then just as quickly, ‘experience a flaw.’ Seeing both the power and the limits of these tools builds confidence and reduces anxiety.”
- Learn together, not alone.
- Prioritize transparency.
- “What people want from leadership is clarity: what skills they should be building, what opportunities they have to develop them, and how their work might evolve.”
Software Engineering
GraphQL API Basics and Best Practices
This was a brief refresher about how GraphQL solves certain problems you can run into with REST APIs, namely underfeteching overfetching.
The Industry Is Catching Up: The Only Way to Generate AI Code That Works
- Uncle Bob no longer reads the code his agents write. He uses various types of tests.
- Adam Tornhill has dropped Red/Green/Refactor with an agentic flow. He still drives new changes by a failing test. https://adamtornhill.substack.com/p/practices-i-abandoned-with-agents Specifications are more important.
- Sean Grove at AI Engineering is also focusing on specs.
- Kent Beck calls TDD a superpower.
- Emily Bache said all the tests form the spec.
- Anthropic’s guidance for Claude Code: “If you can’t verify it, don’t ship it.”
- The engineer’s job is to tell the machine exactly what the code should do in a form that can be checked.
- Dave is advocating for ATDD, which is an intent-based approach.
- AI changes the economics of one activity: producing the code. We still have to specify what we want (prompts are more vague than a program), verify that it works (AI can change more than we asked for), then proceed in small, safe steps (humans can’t keep up).
- How fast can we generate releasable systems?
- Tests generated by AI via it looking at code only confirms that the code does what it does, not whether it’s correct. (These are characterization tests.)
- Dave called out that Gojko Adzic who wrote the book on specification by example found SpecKit lacking; the real spec ended up being the tests.
- The flow has always been: vague wish -> story -> example -> specification -> code/tests -> working software.
- For each feature: write the spec, see tests fail (confident red), work in small steps
- Spec-driven development doesn’t mean we stop reading the code, but it’s a choice we make when the risks warrant it.
- “[Specs are] often not very assertion-heavy, giving us a false sense of security.” — Martin Fowler
- Humans write the specs, AI implements the tests, AI implements code to make the tests pass, tests run in the pipeline
- My issues:
- Language is an imprecise form of communication
- AI coding tools are non-deterministic
- Every thing you prompt AI to do is a suggestion
Technology
The New Tesla Demo Captures Why Big Tech is Broken
- Companies are inventing the future instead of trying to improve the present
- Cory Doctorow says these companies want to be seen as growth companies instead of stable companies. Growth companies get much larger valuations from investors. They have to come across that there’s still lots of innovation coming in the future.
- Engagement-pilling. Silicon Valley has been built on foundation of engagement (e.g., social media). People are seeking out disruption as a source of meaning. Futuristic stuff gets more engagement.
- It’s hard to build useful products. The people holding the reins at the AI companies are not brilliant product strategists.
- Not solving current problems moves the risk onto the consumers. These companies create false economic growth because their stock price does not reflect their true value.
- Let R&D labs invent the future.
- Don’t engage with the tech bros on X.