Market snapshot — the page that ages
📖 Walk me through it — plain English
Most of this guide teaches things that will still be true in five years — how a hash map works, how to narrate your thinking, how to design a rate limiter. But job hunting also needs perishable facts: which companies are hiring hard right now, who allows AI tools in their interviews, which models are current. Mixing those into evergreen lessons makes the whole guide rot; quarantining them on one clearly dated page keeps the rest trustworthy. That's this page: the snapshot. Read it once before you start outreach, and re-verify anything load-bearing before an actual interview.
These are the dated figures behind the durable mechanism in the raised-floor lesson — read that for why; these are the current how much. All directional; re-verify before citing.
- Entry hiring fell hard 2023→2024 and stayed depressed — entry-level tech hiring estimated down roughly 25–50% off its peak, with computer-programmer employment at its lowest level since the 1980s.
- Big Tech cut new grads sharply. Google and Meta have hired on the order of ~50% fewer new grads than their 2021 peak. At large public tech firms, the share of workers aged 21–25 roughly halved Jan 2023 → Aug 2025 (Meta: ~15% → ~6.8%).
- CS-grad unemployment is elevated — recent CS grads around ~6% vs a ~4.8% all-grad average (computer engineering higher still). But CS underemployment stays low (~16–17%): the entry door is narrow, the career floor for those who get in is not collapsing.
- The "seniorization" of entry roles is measured. PwC's 2026 AI Jobs Barometer (>1B job ads, 27 countries) found AI-exposed entry-level roles are about 7× more likely to require traditionally senior "human" skills — judgment, leadership, stakeholder management. "Seniorized" entry roles grew ~35% since 2019 while other entry roles shrank ~10%.
The tier shapes are durable and live in the target list lesson; these are the current names occupying each tier.
Anthropic, OpenAI, Google DeepMind, Cohere, Mistral, Hugging Face, Together AI, Replicate, Fireworks.
Cursor, Lovable, Bolt, v0, Granola, Cresta, Sierra, Decagon, Cognition, Adept, Harvey, Glean, Perplexity, Linear, Notion (AI features team), Intercom (Fin team).
Palantir (the original), Anthropic Applied AI, OpenAI Forward Deployed, Sierra, Decagon, Scale AI, Anduril.
Notion AI, Intercom Fin, Zendesk AI, HubSpot, Asana, ClickUp, Datadog, Snowflake Cortex, Atlassian Rovo.
YC company directory filtered by newest batch + AI tag. Fastest hiring decisions, often under 2 weeks.
Banks (Capital One, JPMorgan, Goldman tech), healthcare (Epic, Cerner, Veeva), big enterprise (Salesforce, ServiceNow, Workday).
The industry has split. Confirm per role/stage before the screen — policies move fast. The AI collaboration lesson teaches the durable workflow either way.
- Canva (since Jun 2025) — requires AI; failing candidates lacked judgment to guide it, not coding skill
- Meta — AI-assisted coding round on CoderPad (Oct 2025+)
- Shopify — your own IDE + AI tools over screen share
- Google — "human-led, AI-assisted" format (Gemini in the room) that scores AI fluency broadly: prompt engineering, output validation, debugging
- Sierra, Vercel, Linear, Perplexity for AI-product roles
- Anthropic — explicit per-stage candidate AI guidance; not always allowed
- OpenAI — work trial allows tools; live screens often not
- Cursor — live coding screen without AI; onsite uses their editor
- Most YC startups — founder discretion
- Amazon — AI tools DQ'd in the coding screen (do-not-hire on detection), yet a separate GenAI-fluency round now probes AI/ML understanding (often blended with DSA): banned in the round, expected as knowledge
- Most HFTs (Citadel, Jane Street, Two Sigma, HRT) — final rounds onsite-only
- Apple, Microsoft, Cisco — onsites returning specifically to defeat AI cheating
The names you'll see preloaded in AI-assisted rounds and discussed in AI-product interviews: GPT-5.x (OpenAI), Claude Sonnet 4.5 / Opus 4.x (Anthropic), Gemini 3 Pro (Google), and open-weight Llama 4 (Meta) for self-hosting. These rev constantly — by the time you read this the point release has almost certainly moved (the frontier was already past GPT-5.5 / Opus 4.8 / Gemini 3.1 mid-2026). Each vendor ships a flagship / workhorse / cheap tier; the tier shape outlives every rename. When a lesson names one of these, treat it as "a current frontier model," not a permanent fact, and check the vendor's own page for today's exact version.