COMPANY PACK · TRANSPARENCY

✓ VERIFIED · 4 PUBLIC SOURCES

Anthropic

Applied AI / Eng · frontier-lab · One curated set a day, tuned to this loop

Updated 2026Q2

Focus areas ranked by how strongly they show up across independent public sources. Tags shift by role & level — refreshed quarterly, never a guarantee.

MOST-TESTED FOCUS AREAS

✦

Agentic AI / AI-assisted

HIGH · 4 sources

Agentic infra for Claude Code · tool-calling on MCP, eval suites that surface safety regressions

100%difficulty

▥

Test Architecture

MEDIUM · 3 sources

System-design round = build an eval harness (not RAG) · measure whether the model actually helps the user

82%difficulty

🔍

Scenario-based

MEDIUM · 3 sources

Reason through agent trajectories, tool-call errors, hallucination and safety cases

60%difficulty

</>

Hands-on Coding

MEDIUM · 2 sources

Coding + ML fundamentals with production accuracy

45%difficulty

ALSO TESTED, LIGHTER — BELOW THE TOP 4

⚖Bar-raiserSafety-mission alignment matters but is not a standalone veto round in public accounts.

High 5+ · Medium 3–4 · Emerging ≤2 sources

One curated set a day, tuned to this loop. Unlock to start.