Most important take away

The path to the top 1% of engineers (E7+ at Meta) requires obsessive impact-driven work, but pure “coding machine” output has a ceiling — beyond senior staff, advancement depends on developing taste, judgment, and broad influence. In the AI era, this taste-and-judgment gap is the central crisis: junior engineers can produce volume with LLMs but lack the war-scars-driven judgment that distinguishes senior engineers, and the industry has no scalable solution yet for transferring it.

Summary

Actionable insights and career advice from Michael Novati (ex-Meta E7, founder of Formation):

• Drive impact over checkboxes. Novati reached E7 in six years by giving his all to Facebook’s mission, not by chasing comp or promotion criteria. Promotions followed naturally from outsized impact.
• Skipping meetings is only acceptable when the alternative is more valuable. He bypassed meetings only when working with PMs in advance produced better outcomes — never to avoid politics for personal convenience.
• Recognize ceilings honestly. “Coding machine” archetypes hit a ceiling around E7 because higher levels require influencing thousands of engineers, which is impossible as a lone-wolf IC. Know whether you actually want broad influence before pursuing it.
• Less-than-meets-expectations reviews can be informative. His first sub-par review wasn’t about output but about needing broader influence to grow. He used it as a signal that the next level wasn’t aligned with his passion and chose to leave.
• For breaking in without traditional credentials, two pieces of advice:
  1. Take a tech-adjacent job that leverages a domain you already have taste in (e.g., golf software if you’re a golfer), then transition into engineering internally.
  2. Start a real company — even a tiny LLC with a Stripe account and 10 paying customers builds the taste/judgment that gets you hired.
• Long-term, lean toward apprenticeship-style models (e.g., University of Waterloo’s co-op program with six internships across five years). New grads from those programs hire in at near mid-level.
• AI is amplifying senior engineers, not juniors. A K-shape is emerging: those with taste/judgment multiply output via AI agents; those without struggle to know what to build. “A really good solution to the wrong problem is worse than doing nothing.”
• AI prompts are getting more compressed. Novati now codes via terse pseudo-language (“persona class, red squares, make blue when…”) because he already had the codebase mentally mapped — AI removed the typing bottleneck.
• Bootcamp economics are broken. Top engineers earn 7–9 figures, so bootcamps end up staffed by recent grads — no taste-and-judgment delta, no apprenticeship value. Worth watching: Mercor and Scale AI paying senior engineers ~$1,200/hr to train LLMs to capture that judgment.
• Coding interviews are “mid, not bad.” Defend leetcode-style interviews — they test thinking process, not memorization. The signal calibrated interviewers seek is taste and judgment under varying constraints, not the optimal big-O.
• Meta is rolling out AI-augmented coding interviews (starter code, larger codebases, simple LLM access) alongside whiteboarding. Expect this to spread; prep fundamentals (communication, thinking aloud) don’t change.
• The top engineers aren’t on LinkedIn. Companies actively protect them; recruiter relationships are built privately over years. Compensation philosophy at Meta: pay engineers $1 more than they think they’re worth — efficient and retentive. (Caveat: this is the legacy demographic; younger engineers may legitimately mix engineering and personal brand.)
• Be vulnerable about failure. Taste and judgment come from screwing up — Facebook’s SEV reviews and Amazon’s COE docs are gold mines. Senior engineers should be more open about mistakes so juniors can learn.
• “Vibe coded million-dollar SaaS in a weekend” is overrated. Real businesses need SOC 2, legal, insurance, support — chasing money over building real skill is a trap. Find what you’re genuinely good at and create value from it.
• AI’s productivity gains likely flow to more output, not more leisure, because of the existential pressure on jobs.
• Founders must be willing to accept their company shouldn’t exist if it stops adding value. Formation may not exist in 5 years if interview prep changes radically — and that’s fine.

Tech patterns mentioned: React migration (early days at FB), Code Health Summit (internal cross-team migration showcase), SEV review process (Facebook), Correction of Errors / COE (Amazon), tree traversal / BFS by level (real on-the-job use case in DOM rendering), small-group mentorship with calibrated level-deltas (Formation’s model).

Chapter Summaries

• From intern to E7 in 6 years. Novati attributes the climb to dedication and impact-driven work, not box-checking. The hardest jump is E6 to E7 — only ~1% of the company reaches it, and there’s no clear template.
• Opting out of meetings (with caveats). He bypassed meetings only when alternatives produced more impact, often coordinating with PMs beforehand. Not a recommended general strategy.
• The coding-machine ceiling. A lone-wolf IC profile hits a ceiling around E7 because higher levels require influence over thousands of engineers. Novati left Meta in part because pursuing E8 required becoming someone he didn’t want to be.
• Code machines in the AI age. AI removes the typing bottleneck for engineers with mental maps of the codebase; Novati’s commit volume doubled after GPT-4. The dry-code principle is loosening when AI manages the code.
• The taste-and-judgment crisis. New engineers can’t develop judgment because LLMs do the foundational work that used to build it. CS programs and bootcamps are in upheaval.
• Apprenticeship as the answer. Waterloo-style co-ops produce near-mid-level engineers as new grads. Bootcamps fail at this because economics prevent hiring senior instructors.
• Two pieces of pragmatic advice. Leverage existing domain expertise into a tech-adjacent role; or start a real company with paying customers to build judgment.
• Defending the coding interview. “Mid, not bad” — the spirit (test problem-solving and code organization, not framework memorization) is sound, even if execution varies. The harder issue is interviewer training, especially at scale.
• Meta’s AI interview rollout. Practical interviews with starter code and LLM access supplement (not replace) whiteboarding. Expect iteration and other companies to follow.
• The top engineers aren’t on LinkedIn. Backchannel recruiter networks place the highest-paid engineers; companies actively protect them from public visibility. Compensation philosophy: pay $1 more than they think they’re worth.
• Overrated: “vibe-coded SaaS in a weekend.” Real companies need SOC 2, legal, support — chasing money without skill-building leads to dead ends.
• AI, jobs, and the future of Formation. AI productivity likely flows to more output, not leisure. Founders must accept their company shouldn’t exist if it stops creating value.