The new GPT-5.6 family: Luna, Terra, Sol

Simon Willison · 2026-07-09 · 4 min read

TLDR: OpenAI's GPT-5.6 family (Luna/Terra/Sol) launches with strong agentic benchmark claims and significantly lower pricing than Claude Fable 5, but the headline numbers are muddied by OpenAI simultaneously publishing a critique of SWE-Bench Pro — the one benchmark where Fable 5 decisively beats them (80% vs 64.6%). The most practically interesting additions are API-level features: programmatic JavaScript-based tool orchestration, built-in multi-agent subagent spawning, and explicit prompt cache breakpoints borrowed from Anthropic's model.

From Cloudwashing to O11ywashing

Charity Majors · 2025-11-24 · 8 min read

TLDR: "Observability" has been so thoroughly co-opted by legacy monitoring vendors that senior engineering executives now describe core observability capabilities — correlating app, business, and system telemetry across individual user journeys — as unsolved problems requiring custom tooling, without realizing that's what real observability already does. The fix isn't better technical education; it's reframing the pitch to execs around business outcomes rather than data structures, because they're making billion-dollar purchasing decisions based on a term that now means nothing.

How many pillars of observability can you fit on the head of a pin?

Charity Majors · 2025-10-30 · 10 min read

TLDR: The "pillars of observability" (metrics, logs, traces, etc.) is a marketing construct, not a technical one — the correct term is signals, as defined by OpenTelemetry. Vendors push the pillars framing because it justifies selling you separate, siloed products for each data type, which multiplies your costs and forces engineers to manually hop between disconnected datasets. The better architecture stores all signal types together in unified columnar storage, letting you derive metrics, traces, and logs from the same underlying structured data instead of duplicating it everywhere.

Leading Complexity – Year 4: Navigating the Age of AI (and Everything Else)

Henrik Kniberg · 2025-06-12 · 3 min read

TLDR: Crisp's "Leading Complexity" program is entering its fourth year with a focus on AI as a leadership challenge, not just a technology one. The core argument is that as AI reshapes organizations, leaders need new mental models around human-AI collaboration, decision-making under uncertainty, and adaptive organizational design. The 2025 cohort runs October–December via ten live sessions with speakers from the agile/complexity space.

Why Kotlin Is Better Than Whatever Dumb Language You're Using

Steve Yegge · 2017-05-17 · 26 min read

TLDR: Kotlin hits a rare sweet spot: it's learnable in days, interops seamlessly with Java/Android, and delivers the expressive power of modern languages (Scala, Swift) without the steep learning curve or syntactic overhead. For Java developers frustrated by the language's glacial evolution, Kotlin is the pragmatic upgrade that actually sticks in production.

Fragments: June 16

Martin Fowler · 2026-06-16 · 8 min read

TLDR: Both AI enthusiasts and skeptics are right, and that's the core problem — shipping code faster than engineers can understand it is a genuine existential risk, just as falling behind AI-adopting competitors is. Charity Majors' key point is that there's no shared feedback loop between these camps, so engineers who actually engage with AI in practice — owning both the wins and the costs — are the ones who will earn the credibility to shape how it gets used responsibly.

Unstick Your Stuck Thinking

Kent Beck · 2026-05-01 · 2 min read

Kent Beck introduces "Thinkies" — named cognitive reframe patterns (like "Can't/Because") that help break circular reasoning by flipping the logic of a stuck argument to surface new solutions. For senior engineers who routinely hit organizational deadlocks like "we can't improve X because of Y," these are concrete mental tools for escaping false constraints rather than just accepting them as given.

Parkinson's

Kent Beck · 2026-04-16 · 6 min read

TLDR: Kent Beck has been diagnosed with Parkinson's disease, which he expects to progressively worsen over the next 5-15 years. This has forced a sharp reprioritization he calls "the time value of time" — doing meaningful things now is worth far more than deferring them, so he's refusing high-paying work that would consume his best remaining years. He plans to keep coding, writing, and pursuing his mission, just with fewer compromises.

Ideological Resistance to Patents, Followed by Reluctant Pragmatism

Martin Fowler · 2026-03-05 · 7 min read

TLDR: Software patents are a broken system that entrenches power rather than rewarding innovation — but for open-source startups, ideological opposition isn't a legal defense. When a large player can weaponize broad IP claims against your own features, defensive patenting becomes a pragmatic shield, not a contradiction of principles. The fact that principled open-source builders feel compelled to file patents is itself proof the system is working exactly as badly as critics claim.

How can you not be romantic about programming?

Thorsten Ball · 2020-09-08 · 3 min read

TLDR: The global software stack is an incomprehensibly vast, fragile, multi-generational artifact — code decades old sits alongside code written last week, maintained by handfuls of people, quietly holding critical systems together. Once you grasp the sheer scale of accumulated human ingenuity behind it all, "magic" stops feeling like a metaphor and starts feeling like the only honest description.