Article illustration 1

Amid deafening hype around generative AI, veteran software architect Tim Bray—known for co-authoring XML and engineering roles at Amazon and Google—has published a contrarian manifesto predicting GenAI's trajectory. His core thesis? The economic frenzy is unsustainable, but AI-assisted coding will fundamentally reshape development—just not in the ways evangelists promise.

The Unfixable Hallucination Problem

Bray asserts LLM hallucinations aren't solvable with current techniques, citing OpenAI's own research: "Hallucinations are an inevitable result of current training practices." He argues that connecting model outputs to ground truth remains elusive, and the absence of breakthroughs despite massive investment suggests limits to progress:

"If there were a way to eliminate the hallucinations, somebody already would have. An army of smart, experienced people, backed by effectively infinite funds, have been hunting this white whale for years."

This has profound implications: GenAI cannot reliably replace human judgment in knowledge work without producing error-riddled "workslop" that erodes quality and brand trust.

The Myth of the Job Apocalypse

While AI aims to replace knowledge workers, Bray predicts mass layoffs won't materialize. "Reverse centaur" models (AI doing work, humans fixing errors) create productivity gains offset by cleanup costs. Meanwhile, "centaur" approaches (humans guiding AI) improve quality but not sufficiently to justify workforce reductions at scale. Recent studies showing modest or negative productivity gains from AI tools bolster his case.

The Unsustainable Economics

Bray echoes Cory Doctorow's warnings of an impending "economic AI apocalypse," pointing to Deutsche Bank analysis that current spending requires "parabolic" growth to sustain. He notes grimly:

"The only people making money are those selling gold-mining equipment to the peddlers... This cannot go on forever, so it will stop—probably by 2026."

However, he diverges from doomsayers, predicting fallout will primarily hit investors rather than triggering systemic economic collapse.

Code Generation: The Exception That Proves the Rule

For developers, Bray sees a nuanced future:

Where AI will thrive:
- Application logic (e.g., "Depreciate values in the AMOUNT field")
- Boilerplate (Android/AWS API integrations, CSS layouts)
- SQL queries and StackOverflow-style lookups

Where it will struggle:
- Low-level infrastructure (memory/algorithm optimization)
- Concurrency models requiring deep systems knowledge
- Human-centric interaction design

Critically, he notes code-generation success hinges on rigorous testing frameworks that validate outputs—making test suites more vital than ever. "The quality of help you get depends on your test framework. Which warms my testing-fanatic heart."

The Ethical Reckoning

Bray reserves sharpest criticism for GenAI's ethical rot: "It’s being sold by a panoply of grifters and chancers... who know their dream world would be generally shitty." He highlights environmental costs and exploitative labor practices in AI supply chains as unresolved scandals.

His ultimate prediction? After the bubble bursts, we'll retain useful coding tools while discarding the dystopian workforce replacement fantasies—a future where "we won’t have to live in the world they imagine."

Source: Tim Bray's GenAI Predictions