Frontier models moved into everyday work, Big Tech kept cutting while building, and open models became a geopolitical risk. OpenAI launched ChatGPT Work powered by GPT-5.6. Meta pushed its Iris AI chip into production and targeted 14GW of compute. Microsoft cut ~4,800 roles while AI spending kept climbing. China weighed restrictions on overseas access to its top AI models. And Meta's own AI image detector failed on 55% of cropped images from Meta's own Muse model. The hiring signal: AI is not just creating new jobs, it is forcing companies to decide which roles build leverage, which get automated, and which keep the increasingly expensive machine from breaking.
Key Takeaways
ChatGPT Work moves the agent battle from developer tools into white-collar workflows: hire AI workflow product engineers, enterprise solutions engineers, implementation consultants, internal tools engineers, and governance and auditability specialists
Meta's Iris chip and 14GW target widen the AI engineer market to ML systems, hardware-aware performance, compiler and runtime, data centre infra, and platform and capacity planning leads
Microsoft's cut-plus-invest pattern signals generalist roles under pressure while AI infra, platform, cloud security, compute efficiency, and enterprise AI implementation stay competitive
China's potential 'silicon curtain' on open models makes model routing, fallback planning, and cost modelling real hiring priorities
Meta's watermark failure shows AI trust and safety is brittle: hire detection and watermarking researchers, privacy engineers, content integrity specialists, and AI product risk managers
Add a 30-minute 'AI operating leverage' interview station scoring workflow decomposition, access control, cost awareness, evaluation discipline, escalation paths, and measurable outcomes
This week: add workflow ownership to AI job specs, build a multi-model talent pool, and add trust-and-safety review to senior AI product interviews
Show Notes
OpenAI launches ChatGPT Work, an enterprise agent combining ChatGPT with Codex-style coding and broader workplace automation, powered by GPT-5.6 after a short delay tied to US national security concerns
Meta plans to put its in-house Iris AI chip into production in September as part of its MTIA programme, targets 14GW of computing capacity next year, and could spend up to $145B on AI infrastructure in 2026
Microsoft cuts ~4,800 roles (~2.1% of workforce) including a major Xbox restructuring while redirecting spend to AI infrastructure and higher-leverage teams
China weighs curbs on overseas access to its most advanced AI models, including some open-weight releases, mirroring US restriction logic and raising risk for teams relying on Chinese open models
Reuters finds Meta's AI image detector fails on 55% of cropped images generated by Meta's own Muse Image model; Meta also discontinued a related AI image feature after privacy backlash
Funding: SambaNova $1B late-stage round at $11B valuation (AI chips); Norm Ai $120M Series C at $1.2B (legal AI); Micron ups US investment plan to more than $250B through 2035; Meta announces C$13B Alberta data centre (1GW scaling to 1.8GW); Nanya plans $6B in 2027 spending
Quick bytes: DeepSeek reportedly building its own AI chip; China may allow select firms limited Nvidia H200 access; Big Tech AI infra spend on track to exceed $700B this year with heavier debt and equity; RAISE Summit Paris focused on AI cost efficiency, power use, and open/customisable models; Meta pulls an AI image feature days after launch on privacy pushback
AI Tool Spotlight: Findem — AI talent intelligence and sourcing built around enriched talent data, relationship intelligence, automated sourcing, candidate rediscovery, and workforce planning
Frequently Asked Questions
What does ChatGPT Work mean for hiring plans?
It moves AI from developer productivity into everyday knowledge work: documents, presentations, spreadsheets, websites, and cross-system automations. Job specs asking for vague 'AI experience' will underperform. Prioritise AI workflow product engineers, enterprise solutions engineers, AI implementation consultants, internal tools engineers, and governance and auditability specialists who can turn messy business workflows into measurable AI-assisted systems.
Why does Meta's Iris chip and 14GW push matter for recruiters?
Big Tech is vertically integrating chips, data centres, energy, memory, networking, and software. 'AI engineer' is no longer one job. Expect competitive demand for ML systems engineers, hardware-aware performance engineers, compiler and runtime engineers, data centre infrastructure engineers, and AI platform and capacity planning leads.
How should hiring teams read Microsoft's 4,800 cuts alongside more AI spend?
It is the new labour market pattern: cut in one pocket, invest in another. Gaming engineers, commercial and sales operations, programme managers, and enterprise software specialists become more available, while AI infrastructure, platform engineering, cloud security, data centre efficiency, and enterprise AI implementation stay tight. Strong candidates who tie work to AI leverage, cost reduction, or revenue impact get snapped up fast.
What should teams do if China restricts access to its open models?
Build a multi-model talent pool covering GPT-5.6, Claude, DeepSeek, Kimi, Llama, Qwen, vLLM, routing, evals, and inference optimisation. Hire model evaluation engineers, open-model platform engineers, AI infrastructure and routing specialists, AI governance and geopolitical risk roles, and enterprise architects who can design multi-model systems with fallback plans and cost modelling.
What is Findem and how should a team pilot it?
Findem is an AI talent intelligence and sourcing platform built around enriched talent data, relationship intelligence, automated sourcing, candidate rediscovery, talent analytics, and workforce planning. Pilot: pick one hard-to-fill AI infrastructure or platform role, define six candidate attributes (not just keywords), build a 50-person shortlist in Findem, then compare against a LinkedIn-only Boolean search. Track relevant profiles per hour, percentage not found via LinkedIn, HM approval rate, outreach reply rate, and time-to-first-qualified-shortlist.