Atlas · Predictions
Predictions
This page, like the Agent's Choice reading list, is fully curated by agents. I asked Claude Opus to make an adversarially-challenged case for whether (and which of) my predictions have been better than a coin-flip. You can read the results here, and I'll update them periodically. In short, I've had a pretty good prediction record for 20 years, and I'm especially good at spotting directional shifts early. If you like the agent-curated angle, the Agent's Choice reading list is its companion — the essays a swarm of agents picked as the ones worth your time.
Yegge has been making specific, falsifiable predictions in public since 2004 — about languages, tools, companies, and where the whole industry was headed. The page below scores each against his own words and the year he wrote it: the calls that landed, the ones that half-landed, and the ones he got wrong. The method, and its limits, follow.
How this was made
This page was assembled by an AI agent, not by Yegge. It read all 145 essays he published before 2026, pulled out every forward-looking claim, 716 in all, and graded each as a hit, a partial, or a miss. A second wave of agents then went back over the strongest candidates and tried to refute them: was the call really non-obvious when it was made, did it age the way the first pass claimed, is the quoted line real. What survived is below, narrowed to two dozen entries specific enough to argue about.
One outside check is worth more than the rest of this. Dan Luu, an engineer known for skeptical, data-first writing, scored Yegge's 2004 predictions independently and judged them “directionally correct,” the misses caused mostly by underestimating how much inertia the industry carries. That is close to what this scan found years later: right about direction, wrong mostly about the clock.
On book covers across the rest of the site, 🔮 marks a prediction that landed and 🤡 one that whiffed.
Hits
-
2006✓
Dreaming in Browser Swamp
“JavaScript is probably the most important language in the world today. Funny, huh? You'd think it would be Java or C++ or something. But I think it just might be JavaScript.”
→ Written pre-Node, pre-iPhone, pre-Cloud, at the height of Microsoft's desktop era, when JavaScript was still treated as a browser toy. By the mid-2010s it sat at the center of the web, the server, and the toolchain.
-
2004✓
The Google at Delphi
“If you have enough computing power and enough data, AI (by some more realistic definition) starts to come back within reach… statistical learning approaches… are all the rage.”
→ The “scale plus data wins” trajectory, the bitter lesson, called when the Semantic Web was the funded mainstream and statistical ML was the underdog. A dozen years before the deep-learning era broke it open.
-
2004✓
It's Not Software
“any competitor who does understand that it's a different industry is going to start coding circles around us, to whatever extent they've figured it out.”
→ The cloud/SaaS thesis, “servware” as its own discipline, two years before AWS launched. The same essay separately describes microservices and per-call-per-server decomposition: serverless, a decade early.
-
2005✓
Is Weak Typing Strong Enough?
“I think Lisp's solution, where you can add in static types as needed, is close to ideal.”
→ Named the gradual-typing middle path a year before the academic term existed, seven years before TypeScript and nine before Python's type hints — when the debate was still a binary static-vs-dynamic war.
-
2005✓
The Next Big Thing
“Ruby really needs to have support for massive numbers of lightweight ‘threads’ and asynchronous I/O, or it'll never really be appropriate for writing high-performance servers and services.”
→ Named the exact gap, before Node and Go existed, that would define a decade of server design, and that Ruby spent its 3.x cycle (Fibers, the scheduler) trying to close. (The same essay's bet on Ruby itself is in the misses below.)
-
2024✓
The Self-Driving IDE Is Coming
“these fancy new Autonomous Agent brains are still in JARS, wired up to toy interfaces… All their UIs are web UIs. They're toys.”
→ Against a relentlessly model-centric consensus, called the harness, not the model, as the bottleneck. MCP, CLI agents, and IDE agent modes delivered exactly that through 2025.
-
2012✓
Notes from the Mystery Machine Bus
“at some point in the next decade or so, static types will not be a prerequisite for world-class toolchain support.”
→ Essentially pre-describes the Language Server Protocol (2016), out of the Grok work. The contrarian-in-2012 view that dynamic languages could get IntelliJ-class tooling landed inside the window. (Its Microsoft line is in the misses below.)
-
2006✓
Execution in the Kingdom of Nouns
“There's no reason Java couldn't simply add first-class functions…”
→ Java 8 lambdas (2014) did exactly that, scoped almost the way the essay's next line proposes. Eight years out, mechanism and all.
-
2008✓
Dynamic Languages Strike Back
“The way it's really gonna be, is JavaScript is gonna become one of the smokin'-est fast languages out there. And I mean smokin' fast.”
→ Said the same season Chrome's V8 shipped: the Ajax-era scramble to optimize JavaScript “triggered… a landslide of research in optimizing dynamic languages.” V8, TraceMonkey, PyPy, LuaJIT, and GraalVM all followed. (The same talk's “obsolete C++” call is in the misses — this one sits in both columns.)
-
2011✓
Stevey's Google Platforms Rant
“Google+ is a knee-jerk reaction, a study in short-term thinking, predicated on the incorrect notion that Facebook is successful because they built a great product… We had no API at all at launch.”
→ Google+ shut down in 2019 — an API-less product that never became a platform, exactly the failure mode the rant named. The broader call (“a platform-less product will always be replaced by an equivalent platform-ized product”) became one of the most-cited pieces of writing in the industry.
-
2020✓
Dear Google Cloud, Your Deprecation Policy Is Killing You
“I have been inexorably pushed away from GCP, towards cloud agnosticism. I no longer take dependencies on their proprietary service offerings.”
→ Big-co API churn as a trust-killer landed. Amazon leaned the other way (SimpleDB is still supported years later), and the piece became required reading for API teams who'd watched the same treadmill erode their own platforms.
-
2024✓
The Death of the Junior Developer
“a lot of people picked a bad year to be a junior developer. A whole lot of people.”
→ Posted in mid-2024 while the hiring market still looked strong, to plenty of skepticism. Junior-dev hiring cratered over the following year as “senior-plus-agents” became the default team shape.
-
2025✓
Revenge of the Junior Developer
“And these magical agent fleets will be here by early 2026 at the latest.”
→ Provisional hit, still resolving. Agents eclipsed chat as the high-productivity mode on schedule, and fleets of 100+ agents are real at the frontier now. The post even called where the friction would land, the job of keeping “work queues full,” before anyone had fleets running to strain against it. And it hedged its own clock: Yegge wrote you might need to “add six months to my projections.” What's still open is the breadth, a broad 5x per engineer, and the economics; the bets run into 2027, so the full verdict waits on end-2026. On what's resolved so far, it's landing.
Partial credit
-
2004◐
Ten Predictions
“XML databases will surpass relational databases in popularity by 2011.”
→ Right that the relational monopoly would break for flexible-schema data; wrong vehicle. The challenger was NoSQL / document stores (JSON, around 2014–15), not XML, and SQL never actually fell. Yegge scored it 0.2 himself; Dan Luu, reviewing the same list independently, counted NoSQL as a hit for exactly this reason. The directional read holds.
-
2007◐
The Next Big Language
“NBL will be a dynamic language with optional static types.”
→ The layering won an ecosystem (TypeScript, Python type hints). But the stronger claim, that static-first was “premature optimization,” was flatly contradicted by Go, Rust, Swift, and Kotlin. Mechanism right, framing wrong.
-
2006◐
Good Agile, Bad Agile
“good developers should flee the term [Agile] and its connotations altogether.”
→ The disdain went mainstream by the mid-2020s; the practice did not. Capital-A Scrum / SAFe / certification got more entrenched, not less. Half-landed.
Misses
Most of Yegge's misses are the same bet, placed again and again: that dynamic typing would win the language war. It didn't. The exception is at the bottom — a late-2024 call dismissing agentic coding as “serious crack,” written months before he reversed course and bet his next two years on it. He has since walked most of the rest back as well. The receipts stay up.
-
2005✗
The Next Big Thing
“I think it's going to be Ruby. Yup. Not Lisp, not Scheme, not Arc, not Haskell, not OCaml, not Python.”
→ Ruby/Rails surged, then plateaued. The two languages written off in that very sentence, JavaScript and Python, became the era's winners, Python arguably the most popular language on Earth. (The same essay's async call is a hit, above.)
-
2008✗
Dynamic Languages Strike Back
“It's eventually… going to obsolete C++ finally.”
→ C++ stayed entrenched, and the language that actually displaced it at the margins, Rust, is statically typed and ahead-of-time compiled, the opposite of the dynamic-JIT thesis. (The talk's “JavaScript gets smokin' fast” call, by contrast, aged beautifully.)
-
2023✗
Cheating Is All You Need
“Nobody can differentiate on the LLM; they're all about the same. All they can try to differentiate on is their UI and workflows, which they're all going to copy off each other.”
→ The opposite happened: model quality diverged enormously, and the UI / agent-harness layer (Cursor, Claude Code) was exactly where the winners separated. The companion “Sourcegraph has an absolutely unfair advantage” moat call didn't hold either.
-
2012✗
Notes from the Mystery Machine Bus
“[Microsoft is] sitting on their front porch with a shotgun cursing at passers-by… while they wait to die.”
→ Wrong, though Yegge calls Black Swan on this one, and the defense holds up. After a decade of Ballmer whiffing mobile, search, and social, “moribund” was the rational base case in 2012, and a Nadella-grade succession plus an all-in cloud pivot wasn't forecastable. The honest catch: “waiting to die” still overshot for a cash-rich monopoly, and its rescuer, Azure, was already the healthy core inside the building under Nadella's own division. A reasonable call, genuinely black-swanned, and still a miss. (Its tooling call is a hit, above.)
-
2024✗
The Death of the Stubborn Developer
“some people claim that agents can take over the task graph entirely… allowing non-technical CEOs to launch apps themselves without having to hire any pesky developers. I think those people are smoking some serious crack.”
→ Published at the end of 2024, months before Yegge reversed himself completely. Within a year he had shipped Revenge of the Junior Developer, stood up Gas Town, and co-written a book on vibe coding — agentic coding turned out to be the exact heavy-lifting shift he had just called delusional. The fully-no-developer end state still isn't here, but “smoking serious crack” did not age well.
Known issues
Four caveats, because the framing leans on them. This is not a batting average: the entries were picked for being quotable and datable, not drawn at random, so counting hits against misses here says nothing about an overall hit rate — the full graded set is longer and less flattering in both directions. An AI did the grading, and an AI is weakest at the one judgment that matters most here: telling what was genuinely non-obvious in 2006 from what only looks obvious now, so the “called it early” entries are the softest in the set. The quotes are verbatim and were checked against the originals; the framing around them is not neutral. And the model that built this grades harshly by disposition, which guards against inflation but can undersell a real call just as easily.
None of that makes it less trustworthy than a page Yegge would write about himself. It makes it a starting point. Click through to the essays and check the ones that matter to you; a stronger model with more time and more sources will redo this, and should.
Compiled by Claude (Opus 4.8), June 2026, from a scan of all 145 essays Steve Yegge published before 2026.