Overview of It’s ok that computers are (sometimes) better than us
In this Stack Overflow Podcast conversation, host Ryan Donovan speaks with Charity Majors, co-founder and CTO of Honeycomb, about how AI is changing software engineering, why “immutability” may matter for code the way it once did for infrastructure, and why the future of software may depend more on evals, QA, operations, and governance than on traditional code-centric thinking. The discussion ranges from spec-driven development and “boring software” to the social impact of AI, surveillance boundaries, and what it means to stay indispensable as engineering changes.
Main Ideas and Takeaways
1. Code may need to become more like infrastructure
- Majors extends the “pets vs. cattle” infrastructure idea to code.
- The core argument: software should become more regenerative, testable, and replaceable rather than something humans must deeply understand line by line.
- She emphasizes that the cost of software is increasingly defined by maintenance, not initial creation.
2. AI changes the economics of software production
- Majors says she was skeptical of AI for a long time, but believes the field crossed a meaningful threshold recently.
- Her view: LLMs can now generate code at roughly the level of the median software engineer, which changes how teams can approach development.
- Instead of assuming one correct implementation, teams may generate many candidate versions and use evaluation to select the best one.
3. The durable layer is not the code itself, but the evals
- She argues that the most important artifact in an AI-assisted SDLC may be the evaluation system:
- deterministic tests
- conformance testing
- benchmarks
- LLM-as-judge style evaluation where appropriate
- Evals define what “good” means and can act as the stable contract for code generation.
- This shifts engineering toward a QA/ops mindset, where the system’s behavior matters more than any specific implementation.
4. Software engineering is becoming more about systems than source code
- Majors notes that the code is not the source of truth; the running system is.
- Engineers historically focused on “how things should be,” but resilient software must account for real-world behavior, variance, and operational reality.
- She believes engineers will need to think more like operators: understanding failure modes, dependencies, and what can safely vary.
5. Some software should be allowed to change; some should not
- A major theme is distinguishing between:
- high-level application logic, where some variance is acceptable
- low-level critical infrastructure, where speed or unpredictability would be catastrophic
- She compares this to latency tolerance:
- a web payments endpoint becoming slightly slower may be acceptable
- a router slowing dramatically would not
- This connects to the idea of pace layers: some layers of a system should move quickly, while others need stability.
AI, Society, and Human Value
6. AI should be made “boring”
- Majors repeatedly returns to the idea that successful technology becomes boring in the best sense:
- predictable
- well understood
- governable
- safe to operate
- She argues that AI should become boring through:
- strong governance
- clearer contracts
- responsible deployment
- appropriate limits on where it is used
7. The speed of AI adoption is socially destabilizing
- She is concerned not only about the technology itself, but about the pace and attitude of its rollout.
- A key worry: some people are treating mass disruption as a goal rather than a risk.
- She suggests that society needs time to adjust, and that the conversation should include labor, displacement, governance, and public consequence—not just engineering capability.
8. Computers being better at some tasks should not threaten identity
- Majors argues that it should not be threatening when computers outperform humans in certain domains, just as it wouldn’t threaten her if someone were a better CTO.
- Her more optimistic interpretation: AI may re-humanize people by reducing the pressure to define personal worth through intelligence alone.
- She notes that being a person includes far more than cognition:
- family
- care
- taste
- relationships
- judgment
- values
Governance, Surveillance, and Responsibility
9. The line between observability and surveillance matters
- Asked directly about the boundary, Majors offers a simple distinction:
- observability/monitoring is for machines
- surveillance is for humans
- She stresses that engineering tools for systems visibility can become dangerous when applied to people without clear boundaries.
- Governance and legal frameworks are necessary, even if imperfect.
10. Laws can be a meaningful corrective
- She cites examples like HIPAA as evidence that regulation can create real protections and rights.
- At the same time, she acknowledges the difficulty of building effective legal and ethical constraints around rapidly evolving AI systems.
- Her concern is that the speed of AI development is outpacing society’s ability to govern it.
Career Advice and Practical Implications
11. Become indispensable
Majors’ advice for individual engineers:
- Become an expert
- Master the tools
- Develop strong opinions
- Become someone worth listening to
Why this matters:
- Expertise builds credibility.
- Understanding both strengths and weaknesses of the tools helps engineers decide when AI should or should not be used.
- The best way to stay relevant is to build judgment, not just technical output.
12. Taste, wisdom, and charisma still matter
- She references the idea that intelligence is becoming a commodity.
- What stands out now is:
- taste
- wisdom
- charisma
- judgment
- In a world where code can be generated quickly, the human edge increasingly comes from deciding what matters and what should be trusted.
Charity Majors on Meta and Leadership
13. Meta reinforced lessons about centralized power
- Majors reflects on her time at Facebook/Meta as less about technical lessons and more about organizational ones.
- Her takeaway:
- the company was heavily shaped by one leader’s vision and power
- that model can drive speed, but also creates long-term weaknesses
- She contrasts that with her own leadership style, suggesting she is less authoritarian and more collaborative.
Notable Phrases and Ideas
- “Mutability is the sworn enemy of understanding”
- “The cost of software is defined by maintenance”
- “The source of truth is the system”
- “Boring software runs the world”
- “The most durable part of the system would be the evals, not the code itself”
- “Become indispensable”
Final Thought
The conversation frames AI not as a replacement for engineering, but as a force that will push software development toward better abstraction, stronger evaluation, and more operational discipline. Majors’ central message is pragmatic: AI can be powerful and useful, but only if engineers, companies, and society build the right guardrails—and only if humans remain responsible for the systems, values, and consequences around the code.