Most important take away

A small minority (around 2%) of engineers have figured out how to use AI effectively and are becoming dramatically more productive, while the mainstream uses AI shallowly and risks getting their impactful work reassigned to that 2%. To stay ahead, engineers should treat all current AI tooling as temporary scaffolding, lean into orchestrating agents rather than micromanaging code, and shift their own role up the abstraction ladder (senior → EM → director) as AI capabilities grow.

Summary

Actionable insights and patterns from the conversation:

Career growth tests and patterns

• Use a fast-cadence growth check: ask “what can I do this month that I couldn’t do last month?” Don’t wait for a missed promotion (a lagging indicator). Compress the time horizon further as AI accelerates the pace of change.
• Don’t assume your job is only what was assigned. As a junior, get the assigned work done well and fast, then use the freed time to find real problems (your users’, your team’s, the company’s) and build tools for them. Kun’s early promotions at Microsoft came from a self-initiated query-click analytics tool nobody asked him to build.
• Control scope precisely: deliver a scrappy, useful first version with existing building blocks, get feedback, iterate. Don’t disappear into a 2-week rabbit hole.
• “Panic when things are smooth.” If a project is running itself, you are no longer adding value, so look for the next step-function change (new surface, new strategy) rather than incremental tweaks.
• Follow curiosity and switch when you stop learning. Kun’s moves (MSFT → FB, news → games, IC → manager → IC, FB → MSFT for a 0→1 platform) were all curiosity-driven and produced level jumps.
• Don’t worry about the “spotlight effect” when moving back from manager to IC on the same team. Trust and relationships transfer; almost no one is paying as much attention to your title change as you think.
• At senior+ levels, recognize your leverage in inbound recruiting from former colleagues. Negotiate level (not just comp) on the move; most “lateral” moves are under-leveled because the hiring company is incentivized to offer the minimum.

Management lessons that transfer back to AI work

• Bad reasons to become a manager: “bigger impact,” “scale my scope,” “influence more people.” Those are achievable as an IC. Management is really a support role (grow people, recruit, unblock).
• The most transferable skill from EM work to working with AI agents is learning to lose direct control. Don’t micromanage agent output or tweak every line of generated code; instead write down principles, set up systems, and steer through them.

Building product (and why it gates promotions)

• Talk to customers before you build, not after. The Facebook games platform launched a high-score-only mechanic without talking to game developers; retention was bad. After redesigning around developer feedback (turn-based, real-time, trivia, etc.), it took off.
• Senior-to-staff and staff-to-E7 promos at Facebook came from product/strategic impact (taking the games platform 0→1, then evolving it across Messenger, News Feed, and Facebook Gaming live streaming), not from deeper coding. The mindset shift: “title doesn’t matter — do the most impactful thing for the business.”

AI era — the core thesis

• AI is causing an industrial-revolution-scale shift, but it’s bimodal. Inside companies, ~2% of engineers have figured out effective AI workflows and are massively more productive; the other 98% use AI shallowly. CTOs are increasingly routing important projects to that 2% with generous token budgets, while slow teams that still quote three months for a button rename will lose headcount.
• AI is going to take jobs and create jobs. The total number of builders may grow, but the distribution shifts hard.

Tooling patterns and what to invest in

• AI tooling will not stabilize until the underlying models plateau. GitHub Copilot → Cursor → Claude Code → orchestrators on top of Claude Code happened in roughly three years. Treat every tool as temporary scaffold; expect to rebuild your workflow every ~6 months.
• Don’t over-invest in specific tools or in viral markdown-file tricks (e.g., 60-star gists of prompt files). Invest in the underlying mindset: continuous learning, comfort with losing control, orchestration over hands-on coding.
• “True vibe coders” (don’t understand what’s being built) hit a wall at production scale (security, commerce, real users). Engineering experience still matters — but to orchestrate, not to type.

The role-shift ladder

• AI as intern → human acts as senior engineer (heavy supervision).
• AI as junior engineer → human acts as senior (assign tasks, review outcomes).
• AI as senior engineer → human acts as engineering manager (don’t review code; ask: did we test end-to-end, dogfood, load test?).
• AI as EM → human acts as director (developer productivity for the agent fleet: what’s slowing the agents down, what tooling do they need?).
• Eventually CEO-level orchestration with agents as the org. Get comfortable now with not touching code.

Concrete advice by level

• Juniors: build 100 small things for fun. Breadth across stacks beats one polished portfolio project. Find the mental state where building is your default relaxation. AI-native juniors can leapfrog senior engineers whose value was in deep muscle memory for a single language/framework.
• Mid-level: same as junior, plus aggressively follow curiosity when you stop learning. Don’t side-quest while a main quest is interesting; side quests are a signal the main quest is dead.
• Seniors: don’t make yourself the human reviewer for all AI-generated PRs — you will lose that battle. Instead, identify the recurring issues you catch and turn them into automated checks, lints, or review agents. Lean on your systems-design experience to orchestrate fleets of agents without turning the codebase into mush.

Process changes worth questioning

• PR review itself may be obsolete in its current form. It existed because two humans needed checks and balances on each other; in the new world the human author is already the checks-and-balances over the AI. Spend energy on tools that help the original author confidently merge, not on faster second-human review.
• Push more controls into feature gating, A/B testing, and blast-radius reduction at the architecture level rather than relying on human gatekeeping.

Tech / industry takes

• Bullish on Meta producing a useful frontier model (right people, right investment, open-weights philosophy).
• Bullish on Claude Code as an inflection-point product (analogous to ChatGPT for chat), though defensibility is unclear.
• Bearish-ish on local LLMs: useful and fun, but local hardware will never close the gap with data centers; treat them like home solar — supplementary, not the grid.
• “AI PMs are the hottest role” — bullshit framing. Taste and product judgment are no longer the exclusive turf of PMs; engineers and designers with AI now have the same superpower. Expect the talent stack to collapse toward versatile generalists.

Chapter Summaries

• Leaving Microsoft for Facebook: Kun’s “what can I do this month that I couldn’t last month?” growth test and why a stalled answer prompted the jump.
• Origin story: childhood Photoshop, paper-and-whiteboard programming summer camp, building Counter-Strike and MMO cheats for money, CS degree mostly spent building side projects, Microsoft intern on Bing autosuggest.
• Microsoft full-time: shipping fast on Bing features, then using freed time to build an unsanctioned query-click analytics tool that drove his first promotions.
• Joining Facebook via Bootcamp: starting on News search, realizing the team wasn’t a fit after six months, switching to the Games team after seeing an internal recruiting ad in his feed.
• Facebook Games platform: technical wins (9 GB index compressed to 400 MB) plus the bigger lesson — launching a high-score-only mechanic without talking to game developers tanked retention; redesigning around developer needs unlocked growth.
• Two-year detour into people management: VP VJ told him his stated reasons were red flags, his manager coached him through the interview, he got the role anyway. Learned a lot but disliked the grindy calendar and predictable 10-year trajectory; came back to IC on the same team without ego damage.
• Promotion mechanics: senior → staff via taking the games platform to PMF (including travel and product-decision ownership); staff → E7 via expanding the platform across Messenger, News Feed, and live streaming.
• “Panic when calm”: the personal driver behind jumping back to Microsoft as a partner-level IC to build a games platform 0→1 on MSN, where he had to learn the business-development side he never owned at Facebook.
• Career advice by level: juniors should be excellent at the basics and build many side projects; mid-level should follow curiosity hard and never get comfortable; seniors should recognize their leverage on inbound offers and negotiate level jumps.
• The AI era and the 2% vs. 98%: industrial-revolution analogy, why companies are routing impactful work to the small AI-fluent minority, and the risk of becoming a “rename-this-button-in-three-months” team.
• Tooling instability: why the AI dev tool stack won’t stabilize until models plateau, why specific-tool investment is low-ROI, and why mindset and orchestration skills compound.
• Juniors and mids can leapfrog seniors whose deep muscle memory in one stack is now commoditized; “true vibe coders” still hit a ceiling without real engineering experience.
• The role-shift ladder: as AI moves from intern → junior → senior → EM, human engineers should move from senior → EM → director, judged by agent-org productivity rather than personal code output.
• Rethinking process: PR review may be the wrong artifact to optimize; instead automate the senior reviewer’s checks and reduce blast radius via gating/A-B tests.
• Bullish/bearish/bullshit round: bullish Meta frontier model, bullish Claude Code, bearish local LLMs vs. frontier, bullshit on “AI PMs are the hottest role” — predicts a generalist talent-stack collapse where engineers, designers, and PMs all gain each other’s superpowers.
• What’s next: Kun left Big Tech to be a solo builder and content creator, exploring ideas and sharing the journey.