What Developers Are Discussing: Issue 2
A week that shifted the AI industry: Anthropic passed OpenAI in valuation, Steve Yegge declared technical interviews dead, and the most upvoted Hacker News post argues that code isn't even the point. Five pieces worth your time.
1. Domain Expertise Has Always Been the Real Moat
788 points, 495 comments on Hacker News
Summary
Aaron Brethorst articulates what many have felt but never put into words: the hard part of software development has never been writing code — it's building a mental model of the domain before you write the first line.
Example: a payroll system. The code is salary * hours. Reality is garnishments, pre-tax deductions, what happens when an employee quits mid-payroll period, retroactive raises, overtime laws that vary by state. The person who knows these rules is irreplaceable. The person who can write code quickly is replaceable by AI.
Why 788 Points
The article hit a nerve. 495 comments, most confirming from personal experience: "I spent 6 months understanding German accounting rules and 2 days on code," "our best developer is a former insurance agent who learned Python."
In the era of AI code generation, this is the key distinction: AI can write code from a spec, but AI can't create the spec from a conversation with a client who doesn't know what they want.
Practical Takeaways
- Invest in domain knowledge, not the next framework. Understanding accounting/logistics/medicine is more valuable than knowing Next.js
- Talk to users. Edge cases live in people's heads, not in code
- AI amplifies, doesn't replace domain expertise
2. Anthropic Surpasses OpenAI: $965 Billion Valuation
412 points, 464 comments on Hacker News
Summary
On May 28, 2026, Anthropic closed a $65 billion Series H round. Valuation jumped to $965 billion — surpassing OpenAI ($730B) to become the world's most valuable AI startup. In February 2026, Anthropic was valued at $380B. In three months, the valuation grew 2.5x. Annual run-rate revenue crossed $47 billion.
Why It Matters for Developers
This isn't just VC news. Anthropic is Claude, Claude Code, MCP. Every developer using Claude Code depends on this company's strategic decisions.
464 HN comments, mostly debating whether the valuation is justified. For: $47B revenue at this growth rate isn't a bubble. Against: near-trillion valuation without profitability is dot-com 2.0.
Practical Takeaways
- Claude models will develop aggressively — $65B in investment ensures this
- Pricing may change — investor pressure means stronger monetization
- Ecosystem expansion — MCP, Claude Code, Agent SDK get more resources
- Competition intensifies — OpenAI won't stand idle
3. The Last Technical Interview
260 points, 265 comments on Hacker News
Summary
Steve Yegge — industry legend (35 years interviewing, Amazon Bar Raiser, Google Hiring Committee) — declared technical interviews dead.
His argument is empirical, not philosophical: Google internally analyzed the correlation between interview results and actual job performance. Results were "incredibly depressing." The Hiring Committee once accidentally reviewed packets of their own employees — and rejected two-thirds.
The replacement — "Campfire Model": instead of hour-long simulations, paid work on real company code. Candidates get money and portable credentials regardless of hiring outcome. Companies get both signal and actual work product.
Why It Exploded
265 comments because every developer has experienced interview absurdity: reversing binary trees on whiteboards, algorithmic puzzles unrelated to actual work, rejecting candidates who then go to competitors and build brilliant careers.
AI finishes the job: if a model solves 95% of LeetCode in seconds — what's the point of testing humans on it?
Practical Takeaways
- Portfolio > LeetCode. Real projects on GitHub provide more signal than algorithmic challenges
- Paid trial periods are a trend that will accelerate
- If you're hiring: consider a paid trial (1-3 days) instead of whiteboard interviews
4. OpenRouter Raises $113M Series B
447 points, 234 comments on Hacker News
Summary
OpenRouter — "unified API for all AI models" — raised $113 million. Lead: CapitalG (Alphabet's fund). Co-investors: NVentures (NVIDIA), ServiceNow, MongoDB, Snowflake, Databricks, a16z, Menlo Ventures.
OpenRouter solves a concrete problem: instead of integrating with OpenAI API, Anthropic API, Google API separately — one endpoint, one auth, automatic routing to the best model by price/quality ratio.
Why It Matters
The investor list signals a trend: AI model infrastructure is becoming a separate market. Not everyone will build their own models — most need an abstraction layer. OpenRouter is to AI models what Stripe is to payments.
Practical Takeaways
- One API instead of five — saves integration time when using multiple models
- Automatic fallback — if one provider is down, requests route to another
- Cost optimization — cheap model for simple tasks, expensive for complex
- $113M for a router — confirms AI middleware is a serious market
5. "You Can Just Say It"
395 points, 214 comments on Hacker News
Summary
A short, precise essay about how people defend human labor against AI. The author describes a pattern of constant retreat:
- "AI can never do X" → AI starts doing X
- "OK, but humans do it better" → quality converges
- "Fine, but for stylistic reasons I prefer humans" → style is subjective
- "Well, AI can't do it consistently" → consistency improves
- Move to the next "never"
The metaphor: "scuff marks around the base of the goalpost from constant movement."
Why It Matters
The article isn't about "AI will replace everyone" — it's about intellectual honesty. Instead of endless retreat, the author proposes: articulate what you specifically value about human work — and defend it directly, without evasion.
This resonates with the other articles: domain expertise (Brethorst) and decisions vs code (Yegge). Human value isn't in writing for-loops — it's in understanding why the loop is needed.
Practical Takeaways
- Don't defend "coding skill" — defend your ability to understand problems
- Be honest with yourself: if AI does a task better, use AI and switch to tasks where you're irreplaceable
- Value = domain knowledge + judgment + taste — everything else gets automated
Weekly Trend: Reassessing Value
Last week's trend was "rethinking complexity." This week — reassessing value. What's valuable in a developer when AI writes code? Domain expertise (Brethorst). Making decisions, not solving whiteboard problems (Yegge). Honest assessment of your strengths (noperator). Choosing the right tool from dozens of models (OpenRouter). And scale: Anthropic is worth nearly a trillion — because the market believes AI infrastructure is as fundamental as cloud was 15 years ago.
One thought unites all five: code is commodity. Understanding is not.
Comments (0)