Tech Talent Drop - Edition 35

This week’s hiring brief: Google moved to make TPUs friendlier to PyTorch with Meta’s help, GitHub gave Copilot a memory, and late-year mega-rounds signaled where headcount grows next. Use this edition to tune interview loops for AI literacy, cut friction, and point sourcing at teams with new budgets.

The Drop

Google pushes PyTorch on TPUs with Meta
Google’s “TorchTPU” effort aims to erode Nvidia’s software moat by making TPUs more PyTorch-friendly, with Meta collaborating. Expect demand for engineers who can run PyTorch at scale off CUDA, plus infra folks who can evaluate cost and portability across

Copilot gets a memory
GitHub released Copilot Memory in public preview for Pro and Pro+ users. Teams can persist useful context for the coding agent and code review, which shortens onboarding and reduces repeated explanations. Hiring signal: candidates who can design sensible AI usage policies and evaluate agent output will ramp faster.

OpenAI investment chatter and consolidation of compute
Reports indicate Amazon is in talks to invest about 10 billion dollars in OpenAI. If finalized, expect more partner-engineer roles and deeper cloud plus model integration work across enterprise stacks.

AI Tool of the Week

HiredScore
Enterprise talent orchestration that plugs into your ATS to surface qualified candidates, route jobs, and support internal mobility with explainable recommendations. It is positioned for large orgs that want governance, transparency, and measurable lift without ripping out their core HR stack. Start with one business unit, measure recruiter capacity, time-to-shortlist, and hiring manager acceptance.

Hiring and Interview Insight

AI literacy is becoming table stakes
Collibra’s CEO just said lacking hands-on AI familiarity is a red flag for new hires, even outside pure engineering. For technical roles he expects comfort with agentic tools. Apply this by adding a short, open-book task where candidates must use an AI assistant, cite sources, and explain what they chose not to trust.

Funding Watch

All announced in the last 7 days:

• Databricks — data and AI — $4B Series L at $134B valuation. Hiring: platform reliability, AI apps, partner engineering.

• Lovable — AI coding/builder platform — $330M Series B at $6.6B valuation. Hiring: applied ML, product, ecosystem integrations.

• Radiant — portable microreactors — $300M Series D. Hiring: embedded, controls, test, manufacturing ops.

• Nirvana Insurance — AI commercial insurance — $100M Series D at $1.5B valuation. Hiring: data engineering, pricing, telematics integrations.

• Exein — embedded IoT security — €100M growth round. Hiring: firmware security, on-device AI, partnerships.

Why it matters
Capital is clustering around AI infra and tools, energy and hard-tech, cybersecurity, and data-rich fintech. Expect requests for platform reliability, data pipelines, integrations, and partner engineering.

Quick Bytes

• TorchTPU hints at a more diversified compute stack in 2026. Bias your infra JDs toward portability and PyTorch expertise beyond CUDA.

• Copilot Memory means less repetition and faster ramp. Add a policy line on what the agent may remember and retention periods.

• Databricks war chest strengthens the AI platform race. Track post-raise hiring in customer engineering, solutions, and research tooling.

What to do this week

1. AI-literacy loop check
   Add a 20-minute task where candidates use an assistant to summarize a design doc and list three verification steps. Score on judgment, not speed.

2. HiredScore pilot
   Connect to your ATS for one BU. Track recruiter capacity, time-to-shortlist, internal mobility conversions, and hiring manager acceptance.

3. Compute portability audit
   Have a staff engineer outline how your PyTorch workloads would run on TPUs, including cost and migration risks. Use it to sharpen 2026 infra hiring plans.

That’s this week’s Tech Talent Drop. Ship one pilot, measure the delta next Monday, and move on.

That’s all for this week’s Tech Talent Drop — stay informed, and see you next week!