Stop being skeptical about AI for development with Charity Majors

Summary of Stop being skeptical about AI for development with Charity Majors

by Gergely Orosz

1h 25m•August 12, 2026
Listen to Episode

Overview of Stop being skeptical about AI for development with Charity Majors

This episode features Gergely Orosz speaking with Charity Majors, CTO/co-founder of Honeycomb, about how AI is changing software development, observability, reliability, and engineering careers. Charity describes her shift from skepticism to cautious acceptance: AI is now useful enough that the real question is no longer if engineers will ship code they haven’t personally read, but when and under what guardrails. The conversation emphasizes that AI can unlock major productivity gains, but only if teams strengthen validation, testing, telemetry, and production feedback loops rather than weakening them.

Main themes and takeaways

AI is changing engineering, but not in a simple “faster is better” way

  • Charity argues that the software industry rushed too quickly to define AI progress as “speed.”
  • Shipping more often only matters if outcomes improve:
    • better quality
    • better reliability
    • better user experience
    • less toil
  • AI is best understood as a powerful but non-deterministic tool that needs discipline, not hype.

The industry is moving toward shipping code that humans don’t fully read

  • Charity believes it is inevitable that engineers will increasingly rely on AI-generated code without reading every line.
  • The important question is not whether this happens, but:
    • what checks exist before code ships
    • how much trust is offset by testing and evals
    • how engineers validate changes they didn’t author directly
  • She frames this as an extension of existing engineering reality: many developers already rely on teammates’ code, libraries, and production systems they don’t fully inspect.

Reliability is under pressure

  • Charity says reliability is getting worse across the industry in many places.
  • The concern is not just AI itself, but how organizations respond to it:
    • removing reliability roles
    • rewarding speed over quality
    • losing connection between code and production consequences
  • She points to this as a real, observable trend, not just skepticism or fear.

Observability, ops, and QA

Production is part of development

  • A recurring point is that production is not “after” development — it is part of development.
  • Charity strongly rejects the idea that the source of truth lives only in the repo.
  • Fast feedback from production is essential for good engineering.

Ops and QA have the right mindset for the AI era

  • Charity argues that software engineers have historically been too snobbish about operations and QA.
  • Ops and QA are concerned with:
    • correctness
    • validation
    • what actually works
    • what happens in production
  • She sees AI pushing engineering back toward these disciplines.

Code review is overloaded

  • Charity breaks code review into distinct activities:
    • deciding whether a change belongs in the product
    • reviewing API and architecture design
    • catching bugs and syntax issues
    • mentoring junior engineers
  • She argues that not all of these belong in “code review” as usually practiced.
  • In her view, the most valuable human work is deciding whether a change is the right product direction, not nitpicking syntax.

Better validation is needed for AI-generated code

  • Because AI is non-deterministic, teams need more:
    • tests
    • evals
    • behavioral checks
    • smoke tests
    • conformance testing
  • Her core metaphor: if AI increases the “trust debt” in code creation, that trust must be rebuilt elsewhere through validation.

What Honeycomb is doing with AI

“Own the loop”

  • Charity says there is no “human in the loop” in the passive sense.
  • The loop belongs to the engineer or team using the tool.
  • Humans remain responsible for:
    • what AI is allowed to do
    • what is accepted
    • what is validated
    • what ships to production

AI norms and values at Honeycomb

  • Honeycomb has been working through practical norms rather than absolutist rules.
  • One important norm: do not send someone anything you haven’t read.
  • If it would take the recipient longer to read than it took the sender to generate, it is likely “slop.”
  • Charity stresses respect for coworkers’ time and attention.

AI use should improve thinking, not replace it

  • She is open to using AI for:
    • structure
    • feedback
    • assistance
    • reducing toil
  • She is skeptical of using AI to replace deep thinking or writing.
  • For her, writing is thinking on paper, and that should not be outsourced away.

Career advice in the AI era

Senior engineers and leaders need AI experience

  • Charity warns that having AI experience on your résumé is becoming important.
  • Engineers, managers, and directors who ignore AI risk falling behind.
  • If you are not learning it at work, find a way to get exposure to it.

Middle managers should stay close to the work

  • Her advice for managers who want to remain relevant:
    • go back to being an IC for a while if needed
    • get hands-on with shipping and AI tools
    • don’t wait passively for the market to decide for you
  • She believes middle management remains essential for:
    • sensemaking
    • context-setting
    • collaboration
    • helping teams understand the “why”

Junior engineers will be okay

  • Charity is optimistic about junior talent.
  • The hardest part is not proving juniors have value; it’s that engineering value itself is hard to quantify.
  • She believes juniors will thrive if companies continue to give them opportunities, especially through internships and low-risk entry points.

Leadership, business, and organizational change

Being kind is not enough; you must be good at business

  • Charity says the most effective leaders are:
    • kind, caring humans
    • skilled business operators
  • Good intentions alone do not preserve a team or culture.
  • If leaders do not build a viable business, someone else will impose efficiency on them.

Leadership is becoming more hands-on

  • AI is lowering the barrier for leaders to ship code and stay close to engineering.
  • Charity sees this as a positive shift:
    • leaders should know what a diff feels like
    • they should understand PR flow and production impact
    • they should not be detached from delivery

Companies are shrinking teams and increasing expectations

  • She worries that some companies are using AI as a justification for layoffs and anti-management sentiment.
  • Her view: management is still needed, but it must be lean, effective, and close to the work.
  • The best managers help teams make sense of the work, not just assign tasks.

Observability engineering and the future of tooling

Observability is becoming more important, not less

  • Charity argues that AI pushes the industry toward better observability.
  • Logs and metrics are still useful, but they are often just exhaust.
  • The real value comes from rich, connected telemetry and structured relationships between data points.

New primitives are needed for AI systems

  • She discusses Honeycomb’s newer ideas, such as timeline-style views, for systems where:
    • an interaction can last hours
    • multiple agents cooperate
    • traces need higher-level visualization
  • Traditional traces and dashboards are not always enough for AI-driven workflows.

Her second edition of Observability Engineering

  • She explains that the new edition is almost a full rewrite, not a minor update.
  • It covers:
    • deterministic vs. non-deterministic systems
    • instrumentation
    • understanding production behavior
    • AI and non-AI workflows
    • governance and leadership topics
    • build vs. buy decisions
    • vendor partnerships
  • She positions the book as especially useful for platform, infrastructure, and engineering leaders.

Notable ideas and quotes

Key ideas

  • “The question is not if we will stop reading AI-written code, but when.”
  • “Production is not what happens after development; it is a stage of development.”
  • “There is no human in the loop; you own the loop.”
  • “If AI debits trust from code creation, you have to rebuild trust elsewhere.”
  • “Anxiety and excitement are physiologically similar; the difference is agency.”

Memorable framing

  • AI should be made “boring” through discipline, guardrails, and validation.
  • Teams should tell the full story: wins, costs, reliability tradeoffs, and lessons learned.
  • Good engineering in the AI era is less about resisting change and more about learning to control it.

Recommended mindset and actions

For engineers

  • Learn AI tools now, even if you remain skeptical.
  • Strengthen testing, evals, and production observability.
  • Treat production feedback as part of the development loop.
  • Don’t outsource your thinking.

For managers and leaders

  • Stay close to the code and the business.
  • Use AI to reduce toil, not to erase accountability.
  • Build norms that respect coworkers’ time and attention.
  • Share both AI successes and the operational costs.

For teams adopting AI

  • Start with experiments.
  • Define what “better” means before measuring speed.
  • Use AI where it improves outcomes, not just throughput.
  • Keep humans responsible for the loop, the judgment, and the final decision.

Closing thought

Charity’s overall message is pragmatic rather than pro- or anti-AI: AI is real, useful, and not going away, but it should be constrained by engineering rigor. The teams that win will be the ones that combine AI’s speed with stronger validation, better observability, and a healthier relationship between code and production.