How AI is transforming work at Anthropic — 3 patterns × 6 animated, interactive charts
Uses the article’s reported anchors (survey + interviews + Claude Code telemetry) and deterministic synthetic micro-data where needed.
Before = ~12 months earlier (survey: 28% usage, +20% productivity). After = “now” (59% usage, +50% productivity). For telemetry, “Before” ≈ Feb 2025 and “After” ≈ Aug 2025.
Survey: usage 28% → 59%
Survey: productivity +20% → +50%
New work: 27% wouldn’t exist otherwise
Telemetry: tool-call streak 9.8 → 21.2
Telemetry: human turns 6.2 → 4.1
Telemetry: complexity 3.2 → 3.8
① Engineers are getting a lot more done, becoming more “full-stack”
Focus on breadth (cross-domain confidence + task-mix drift), not “hours saved”.
Includes xenographic metaphors: passports, alluvial “career drift”, block-stacking workbenches.
Skill Bloom (xenographic radar morph)
Metric: “Can I successfully ship in this domain?” (0–100), especially outside primary expertise.
morph
AfterBefore
Synthetic, but grounded in the article’s qualitative claims. Hover points for “why this domain expands”.
Team × Task Heatmap (full-stack drift)
Metric: share of a team’s Claude Code usage spent on tasks outside their “core”.
morph
Includes several article examples (Security: code understanding; Alignment/Safety: front-end; Non-technical: debugging + data science).
Usage vs Productivity (132 respondents, constrained)
Anchors: mean usage 28%→59%, mean productivity +20%→+50%, ~14% power users > +100% (article).
Hover a dot: see the individual before→after “jump”. Deterministic synthetic data.
Autonomy + Oversight (telemetry bars)
Anchors: complexity 3.2→3.8, tool-call streak 9.8→21.2, human turns 6.2→4.1 (article).
“Tighter loop”: fewer human turns, longer action chains, and higher complexity.
Loop Latency Violin (distribution morph)
Xenographic: the whole org’s loop-times collapse from “weeks” to “hours”.
Novel cue: density “breathes” faster when loops are shorter.
Iteration Spiral (turns per day)
Metaphor: same calendar day → more learning cycles. The spiral tightens + speeds up.
speed
A “feel” chart. Hover the comet for implied “cycles/day”.
Prompt→Ship Token Flow (pipeline kinetics)
Tokens represent iterations. After = faster tokens + fewer backtracks.
Good for storytelling: “the org becomes less viscous.”
③ Engineers are tackling previously-neglected tasks
Focus on unblockable backlog (work that “wouldn’t have been done”), papercuts (8.6% of Claude Code tasks),
and how the delegation boundary changes what gets attempted.
Backlog Iceberg (27% “new work” revealed)
Anchor: 27% of Claude-assisted work “wouldn’t have been done otherwise” (article).
Metaphor: AI lowers the “waterline” (activation energy), so buried tasks surface.
Papercut Compounding (cumulative curve)
Anchor: 8.6% of Claude Code tasks are “papercut fixes” (article). Minutes mapping is synthetic.
engineers
Small fixes don’t just save time once—they save time repeatedly.
Delegation Frontier (verifiable × stakes)
Based on the article’s delegation heuristics (“easily verifiable”, “low stakes”, “well-defined”).
Boundary expands, but “high-stakes + low-verifiability” remains sticky.
Long-Tail Completion (Pareto tail lights up)
Metaphor: AI turns “nice-to-haves” into “actually done”.
energy
Slider = activation energy drop. Watch the “tail” flip from grey to green.
Maintenance Debt Mountain (area reverses)
Xenographic: “debt” is a landscape—AI makes it cheaper to terraform.