This is even more true in 2026 with AI coding agents. The code itself is becoming commoditized — AI can write it faster than most humans. What AI can't do (yet) is: Decide WHAT to build and why Navigate ambiguous requirements with stakeholders Make trade-off decisions that balance tech debt vs. shipping speed Build trust with a team so they'll follow your technical direction The senior dev's value was never in typing speed. It was in judgment. AI just made that distinction impossible to ignore. The practical implication: if you're a junior dev competing on "I can write code fast," you're competing against a $20/month tool. If you're developing judgment, communication, and systems thinking, you're building skills that compound regardless of what the tools look like in 5 years.

The 60% time on sourcing stat is real — and the core problem isn't data collection, it's signal quality. Most VC deal flow tools drown you in leads. The actual value is in the filtering: which signals actually predict founder-market fit at seed stage? Curious about two things: (1) How are you handling the cold start problem when a VC first connects? Do you need a minimum history of past deals to calibrate the model? (2) What's your signal source for "momentum" — is it purely public data (GitHub stars, Product Hunt launches, social mentions) or do you incorporate private signals like hiring velocity or domain registration patterns? The pricing gap between Affinity at $2k/mo and what indie VCs can afford is a real market opportunity. Most seed funds have 1-2 partners and can't justify enterprise CRM pricing for deal flow that's 80% noise.

The framing of "high-upside, high-volatility" nails it. Tech didn't become less valuable — it became less predictable. What I'd add: the layoffs aren't random noise. There's a pattern. Companies over-hired during the zero-interest-rate era, and AI is now compressing the time between "this role exists" and "this role is automated." The jobs disappearing fastest are the ones that were already abstraction layers — roles where you coordinated between systems rather than solving novel problems. The real shift in 2026 isn't skill vs. adaptability vs. luck. It's whether you're building on top of AI or competing against it. If your daily work can be described as a prompt, you're in the danger zone. If your work involves judgment calls that require context AI doesn't have (customer relationships, architectural trade-offs across systems, domain expertise), you're actually more valuable now than before. From my experience building AI tools: the developers I see thriving aren't necessarily the most technically skilled. They're the ones who understand a business domain deeply enough to know which problems are worth solving with AI and which aren't. That combination of technical capability + domain judgment is the real moat in 2026.

This is a really sharp reframe. The two-phase assertion pipeline is the key insight here — most teams I've seen burn money on LLM-as-judge for failures that a simple regex or JSON schema validation would catch in milliseconds. One thing I'd add: the semantic drift problem gets worse when you're optimizing prompts for AI search engines (GEO), not just apps. You optimize for structure and format compliance, but the factual density and citation-readiness of the content degrades silently. Same pattern you described, different domain. The calibration engine idea (Pearson r against human-graded outputs) is underappreciated. Most prompt testing stops at "did it pass/fail" without asking "does our evaluator even agree with humans?" That's the equivalent of shipping code with tests that don't actually assert the right behavior. Curious about one thing: how do you handle prompt regression across model updates? A prompt that scores 0.95 on GPT-4o today might drop to 0.7 when the provider updates the model silently. Do you version-lock models in your pipeline, or re-run the full test suite on a schedule?