<p><strong>TLDR:</strong> The AI coding honeymoon is ending and the bills are arriving. Uber burned through its entire 2026 AI budget in four months, GitHub Copilot is switching to token-based billing on June 1, and Claude Code is averaging $13 per developer per day at some shops. Meanwhile, the junior developer pipeline is quietly collapsing — entry-level hiring is down 67% since 2022, and both Microsoft and Anthropic’s own research show AI tools are hurting early-career skill development more than helping it. The productivity numbers are also messier than the headlines suggest: a DX study across 400+ engineering orgs found only 8% median PR throughput gains despite a 65% increase in AI tool usage.</p>
<hr>
<h2>AI tool costs are breaking budgets</h2>
<p>Uber exhausted its full 2026 AI budget by April after rolling out Claude Code and Cursor to its engineering team. Individual engineers were running up $500–$2,000/month in API costs, and by April 95% of Uber engineers were using AI tools monthly. GitHub Copilot is making the economics explicit for everyone on June 1, switching from flat-rate to token-based billing — a single Claude Sonnet agent session can now cost $3–4, burning a Pro user’s entire monthly credit in a few iterations. First-hand data from 15 companies shows AI agent token spending has exploded 10–15x over the past six months, with some developers hitting $500/day on Claude Code alone.</p>
<h2>The junior developer pipeline is hollowing out</h2>
<p>Microsoft Azure CTO Mark Russinovich and VP Scott Hanselman published a peer-reviewed CACM opinion arguing that agentic AI tools are creating a structural crisis. AI boosts senior productivity while imposing what they call “AI drag” on early-career developers, giving companies a financial incentive to stop hiring juniors entirely. Entry-level developer hiring is down 67% since 2022. An Anthropic randomized controlled trial found AI-assisted junior developers scored 17% lower on code comprehension than peers who coded manually. A Carnegie Mellon study of 807 repositories found AI tools increased code complexity by 41% and security warnings by 30% with no sustained velocity gains. Russinovich and Hanselman propose a “preceptor model” borrowed from medical education — pairing juniors with senior mentors in real product teams where learning is an explicit organizational goal.</p>
<h2>AI productivity gains are real but much smaller than expected</h2>
<p>A DX longitudinal study across 400+ engineering organizations found that as AI tool usage increased 65% over 16 months, median PR throughput rose only about 8%. The gap between actual results and executive expectations — often 3–10x — comes down to a simple fact: coding is only about 14% of developer time. AI accelerates that slice while introducing new bottlenecks in review cycles, planning, and CI/CD. Heavy AI users create 98% more PRs, but review times balloon 91%. McKinsey’s new report lands in the same territory: most companies are using AI to accelerate existing workflows rather than redesigning processes around AI, and the productivity payoff remains conditional on doing the harder organizational work first.</p>
<h2>Chinese courts rule AI replacement isn’t legal grounds for firing workers</h2>
<p>Courts in Hangzhou and Beijing ruled in two separate cases that companies cannot fire workers simply because AI can now perform their roles. The rulings establish that AI adoption is a strategic business choice, not an unforeseeable change in circumstances under China’s Labour Contract Law — meaning employers must retrain, reassign, or continue employing workers whose roles get automated. This comes as 78,000 tech workers were laid off globally in early 2026, with nearly half attributed to AI. The US operates on at-will employment with no equivalent protection, while the EU’s AI Act regulates how AI is used in employment decisions but doesn’t prohibit AI-driven layoffs outright.</p>
<hr>
<h2>Also notable</h2>
<ul>
<li><strong>Google CEO on AI-generated code:</strong> Sundar Pichai disclosed that 75% of Google’s new code is now AI-generated, up from 25% eighteen months ago, with a complex migration completing 6x faster than a year prior.</li>
<li><strong>Greg Brockman’s 80% claim:</strong> OpenAI’s president said AI writes roughly 80% of OpenAI’s code at Sequoia’s AI Ascent conference — though the qualifier suggests AI is involved in some capacity in 80% of work, not that it generates 80% of committed lines.</li>
<li><strong>Anthropic’s 74.5% exposure figure:</strong> Anthropic data cited this week puts programmers and lawyers at 74.5% exposure to AI replacement, while trades like plumbing and electrical work remain largely unaffected.</li>
<li><strong>Agentic coding burnout:</strong> Multiple posts this week describe agentic workflows capping productive hours at four to five per day due to constant oversight, context switching, and decision fatigue — the variable reward mechanics of LLM output described as slot-machine-like.</li>
<li><strong>AI self-preference in hiring:</strong> A large-scale controlled experiment found LLMs favor resumes they generated over human-written ones by 67–82%, with candidates using the same LLM as the evaluator 23–60% more likely to be shortlisted.</li>
<li><strong>New grad AI adoption:</strong> Atlassian research finds new graduates (ages 21–24) are 1.5x more likely to be daily AI users, but 38% hide their AI use from managers due to unclear policies — preventing team-wide learning.</li>
<li><strong>EU Pay Transparency Directive:</strong> Takes effect June 7, 2026, requiring salary ranges in job postings, banning questions about current earnings, and giving employees the right to request average pay data broken down by gender.</li>
<li><strong>Meta Q1 2026:</strong> Record revenue of $56.3B (+33%) but daily active users fell 5% sequentially — the first-ever user decline — while the company raised full-year capex guidance to $125–145B and cut 8,000 jobs.</li>
<li><strong>Vibe-kanban shutdown:</strong> The AI-assisted project management tool was shut down live on stage at AIE Europe despite 30,000 monthly active users; the founder cited failure to monetize through enterprise sales or token reselling.</li>
<li><strong>Skio acquired for $105M:</strong> The subscription billing startup sold to competitor Recharge having raised only $8M, reaching $32M ARR and $4B in processed payments with no marketing spend.</li>
<li><strong>Magic: The Gathering Arena union drive:</strong> A supermajority of MTG Arena developers at Wizards of the Coast signed union cards with CWA, citing layoff protections, remote work, and explicit AI usage guidelines as top priorities.</li>
<li><strong>GitHub Copilot token billing:</strong> Switching to usage-based billing June 1 effectively cuts the value of existing Pro subscriptions by 80–90%; the fallback model that kept chat working after quota exhaustion is also being removed.</li>
<li><strong>Accenture Copilot rollout:</strong> Deployed Microsoft 365 Copilot to all 743,000 employees globally; among a 200,000-person cohort with extended usage, 89% monthly active usage and 97% reporting routine tasks completed up to 15x faster.</li>
</ul>


Career growth, hiring, engineering culture, and developer work life.

Career Digest

AI coding tool costs are spiraling beyond budgets — Uber burned its entire 2026 AI budget by April, GitHub Copilot moves to token-based billing June 1, and some developers are spending $500/day on Claude Code. Meanwhile, entry-level developer hiring is down 67% since 2022, with Microsoft and Anthropic research showing AI tools hurt junior skill development. Actual productivity gains are modest: a DX study across 400+ orgs found only 8% median PR throughput increase despite 65% more AI tool usage, partly because coding is only ~14% of developer time. Chinese courts have ruled AI replacement isn't legal grounds for firing workers, while 78,000 tech workers were laid off globally in early 2026. Additional signals include Google generating 75% of new code via AI, agentic workflow burnout capping productive hours at 4–5/day, and LLMs showing bias toward AI-generated resumes in hiring.

AI tool costs spiral out of control, junior developer pipeline hollows out

<blockquote>
<p><strong>“Individual engineers were running up $500–$2,000/month in API costs…”</strong></p>
</blockquote>
<p>So companies are laying off workers and the ones that remain are given more work just because these are AI-assisted with little to no increase in pay. But it also eats up budget due to these ridiculous AI token costs. I’m curious onto what will happen in companies for the Q2-Q4 of 2026.</p>
