The reading order
-
1.
The Pre-Series-A AI Startup Hiring Plan: Who to Hire, in What Order, and Why Most Get It Wrong
The full sequence from founding engineer to fifteen, with the comp framework and the mis-leveled hires to avoid.
-
2.
From One Engineer to Fifteen: What Co-Founding Taught Me About Engineering Leadership
What changes at each scale boundary. The shift from delivery to leverage to org design.
-
3.
How to Manage a 4-Person Engineering Team Without Becoming a Manager
The rituals that work at this size — 1:1 cadence, planning loop, decision-making — and the failure modes when you outgrow them.
-
4.
Trust Is the Operating System
The unit is a pod of three to eight people — past eight you stop knowing who's actually stuck and start reading status updates instead of people — so when the team outgrows one pod you repeat it rather than stretch it. Agents run the same operating system as an intern you shouldn't trust off the bat, minus the one part that doesn't transfer: with a person you protect the struggle because that's how they grow, and every agent mistake is a rep for you, not for it.
-
5.
A Scared Team Is Your Biggest Attack Surface
A team afraid to report extends every incident by exactly the length of that fear — detection only starts the dwell-time clock once someone tells it to. What breaks a team isn't a strict manager but an unpredictable one, where the same mistake gets a shrug on Tuesday and a public dressing-down on Friday. Blameless postmortems are incident-response infrastructure.
-
6.
How I'd Hire a Staff Engineer at an AI Startup
The interview loop, the failure modes, the senior-vs-staff calibration that produces correctly-leveled hires.
-
7.
The 'Smallest Possible Slice' Heuristic for Shipping Complex Features
The leverage move that defines senior IC work. Smaller than MVP. Earns the next slice.
-
8.
Every Insane Codebase Was Once Rational
A new grad about a year in — running an entire project plus a younger developer reporting to him — asked how he'd know his architecture decisions were right. Every insane codebase is a perfectly sane response to constraints that existed when it was written, so make the decision that's right for the next three to six months. The exception is one-way doors: data model and storage semantics, your public interface, authentication and identity, anything touching money or compliance-relevant records.
-
9.
The Perfect Hire Is Killing Your Team
The whiteboard interview rewards memorization and pedigree — the weakest predictors of real performance. Hire for trajectory instead.
-
10.
We're About to Stop Making Senior Engineers
The path to senior ran through the grunt work AI now does in one prompt. How to grow architects when the ladder's bottom rungs are gone.
-
11.
The Onboarding Playbook for an Agent-Heavy Codebase
Week 1 is read before you write, authorship-aware; week 2 is a small bounded first PR carrying a one-paragraph how-I-built-this note; weeks 5–8 are the real test — reviewing someone else's agent-authored diff, not writing one. Plus the two layers of convention, the one a linter enforces and the one held only in a senior's head, and why a mentorship plan built on boilerplate tickets hands a new hire work the agent would have done in the same PR anyway.
-
12.
Performance Reviews When Agents Do the Typing
Commit volume died as a proxy — 4,154 commits in six months of an agent-heavy workflow on one project, more than the previous four years combined. What replaces it: the calls an engineer makes about what not to let the agent do, review quality on work they didn't type, problem selection, verification discipline. Plus the five-question 1:1 script to run in the room instead of pulling up a dashboard.
-
13.
A Professional Owns the Whole Outcome
Professionalism isn't process or titles. It's owning the whole outcome — the cost, the failure, the 3am page, the wrong call.
-
14.
The Career I Built on Work Nobody Wanted
The filter was never tedium — it's blast radius: does this force me to understand everything downstream of it? Lavender's security program from zero to SOC 2 Type II, the Cognito-to-Auth0 migration covering 225K+ users and $400M+ in assets, InsideTrack's ten services consolidated down to six. Named as a trade, not a hack: years where the visible artifact of your quarter is a checklist.
-
15.
Prove the Return or Don't Spend the Time
Time is the one budget you can't refill. Every tool and hour has to show a demonstrable return, or it's a tax on attention.
-
16.
How I Triage a New Codebase in 90 Minutes
The 7-step protocol for a useful POV on an unfamiliar codebase in 90 minutes — readme, deps, git log, tests, hot files, auth, incidents.
-
17.
What I Check When I Inherit a Vibe-Coded App
The intake order when there's no living person to ask why: dependency manifest before anything else, auth and secrets second, what the tests actually assert rather than how many files pass, the deploy path, and the data model last because it's slowest to audit. GitGuardian's 2026 State of Secrets Sprawl report found Claude Code-assisted commits leaking secrets at a 3.2% rate against a 1.5% baseline across all public GitHub commits. Full rewrite is the answer far less often than instinct suggests.
-
18.
How We Cut $350K From Cloud Spend in 6 Months (And What I'd Do Differently)
A real GCP-to-Azure migration: lift-and-shift first, optimize after. The $350K saved, the $50K mistake, and when not to.
-
19.
Your Junior Dev Leans on the Agent. Good.
Telling him to use AI less is advice with an expiry date, spent in front of someone who'll watch it get overturned. The readability bar was a proxy for edit cost and that cost collapsed; what's left is verification, falsifiable in a way "is this readable" never was. Break the implementation on purpose and confirm the suite goes red; make the test come from the requirement, not the pass that wrote the code. The self-attack is Faros AI's 22,000 developers — incidents-to-PR up 242.7%, churn up 861%, review time up 441.5% — and none of those three lines is an inelegant abstraction.
-
20.
The Office Mandate Is a Measurement Failure
Nobody mandates presence when they can see output, so the four-day mandate is an admission written in real estate. Agents severed the proxies outright — 4,154 commits in six months into a codebase that grew to about 1.5 million lines, counted honestly as surface area, with hands that did not get faster. Boris Cherny starts batches from his phone and ships dozens of PRs a day; no badge reader would report him as working. Two questions replace the dashboard, and the second is the one people skip: was the goal ambitious enough. Two clean quarters with no near-misses is a management failure, not health.
-
21.
Estimating Client Work When Agents Do the Building
The same ticket is twenty minutes or three days and you can't tell which until you're inside it, so the fix isn't a smarter estimate, it's a different contract. Ranges run 2x to 4x instead of the 20–30% pad hand-written work took. METR's 2025 trial is why you don't quote off vibes: experienced developers took 19% longer on real tasks while forecasting a 24% speedup and still believing afterward they'd been faster. Generation compressed and review didn't — time to first PR review up 156.6%, bugs per PR up 54% — so the number is really pricing judgment and blast-radius control.
-
22.
What Reverse Acqui-Hires Actually Diligence
Microsoft licensed Inflection for roughly $650 million and hired Suleyman; Amazon paid Adept a reported $330 million plus a $100 million retention pool for about 80% of the team; Google took a $2.4 billion non-exclusive license on Windsurf after OpenAI's $3 billion acquisition collapsed inside 72 hours. The buyer isn't grading the repo — the reviewer maps, file by file, which humans hold each critical system, cross-referenced against exactly who is on the offer list, and inventories the IP that isn't code: eval harnesses, data-rights provenance, prompt scaffolding. Which is why the last-minute cleanup sprint is effort spent on the wrong exam.
Other topic guides
- AI engineering What production AI engineering actually looks like in 2026 — the autonomy ladder for agents, the workflow shift, telemetry, and team sizing.
- Security for startups A guided reading order for startup security — what to read first on SOC 2 as a revenue tool, vCISO hiring, and securing AI-native products.
- Elixir and the BEAM for AI systems Why language choice matters for AI systems — BEAM concurrency for agents, what Go frameworks cost you, and why Elixir is the language AI writes best.