Amazon retires Mechanical Turk; GPT-5.6 nears launch

Amazon will stop accepting new Mechanical Turk customers on July 30, putting the 21-year-old crowdsourcing marketplace into maintenance-only mode with no new features planned. Launched in 2005, the platform paid workers pennies to label data and judge content — the invisible human labor that trained a generation of AI systems. The irony is direct: one analysis found up to 46 percent of its workers were already using large language models to do the tasks, so the marketplace that fed AI is now being made redundant by it.

OpenAI is moving GPT-5.6 toward broad release, with reports pointing to a mid-July general availability after a government-gated preview limited to about 20 partner organizations. An OpenAI engineer indicated the top 'Sol Ultra' tier — a mode that coordinates subagents to tackle complex work — is coming to Codex, the company's AI coding agent. GPT-5.6 splits into three durable tiers, named Sol, Terra, and Luna, spanning flagship reasoning down to fast, low-cost inference.

Crusoe, the US data-center builder that constructs AI compute for OpenAI, Meta, and Oracle, is in talks to raise about $3 billion at a $30 billion valuation — roughly triple its worth from a year ago. The company, which pivoted from crypto mining to AI infrastructure, says its development pipeline now tops 40 gigawatts, including a 1.2-gigawatt cluster it is building for OpenAI in Abilene, Texas. The round underscores how much private capital is still flowing into the physical backbone of the AI buildout.

In a widely-shared analysis, investor Tomasz Tunguz argues AI compute spending is on track to rival — or exceed — engineer salaries by 2029. He notes Anthropic already spends about 2.3 times its payroll on compute, roughly $515,000 per engineer per year, while even the top 1 percent of software firms spend about $89,000 per engineer. If the broader market follows that curve, the cost of the AI tools could pass the cost of the people using them — a striking reframe of what "productivity" actually buys.

A new controlled study put coding agents to a careful test: does messy code slow them down? Across 660 trials using Anthropic's Claude Code on matched clean-versus-messy codebases, the agent's success rate barely moved — but on cleaner code it used 7 to 8 percent fewer tokens and revisited files 34 percent less often. The takeaway for teams building with AI: tidy code won't make an agent more capable, but it does make it cheaper and more efficient to run.