Autonomous Agent

StatusEmerging

SignalValidated

EvidenceGrade B

Claude Code

C▼ -10 vs last quarter

Terminal-native agentic coding at scale with dominant enterprise adoption, now re-baselined on Claude Opus 4.8 (May 28, 2026): SWE-Bench Pro 69.2% category-leading, 4x fewer unflagged code flaws, Dynamic Workflows running hundreds of parallel subagents per session.

Rate this tool Add to compare

Dimension breakdown

Score · confidence

UX / DX

79% conf59

Capability

79% conf65

Reliability

79% conf40

Value

50% conf85

Community

79% conf0

Enterprise / Compliance

50% conf65

Autonomy

50% conf60

Integration

50% conf75

Sources blend review platforms, community sentiment, the Signal Radar and practitioner ratings. Scores re-blend each quarter.

Validated maintained: 70% Fortune 100, 8 of Fortune 10, NYSE/Stripe/Ramp/Netflix deployments, overtook ChatGPT in US business AI payments.

Caution

reliability-complaints cap held (autonomy 12) — Opus 4.8 is day-zero with no production-validation window, and the March–May throttling backlash, though actively remediated (5-hour limits doubled May 6, rate limits reset May 15), is not yet a closed 90-day reliability cycle.

Three command-injection CLI CVEs (CVE-2026-35020 CVSS 8.4; CVE-2026-35022 9.9 in CI/CD) remain vendor-dismissed as Informative; CI/CD remote-exploit path is a watch item.

New

Cowork desktop agent (Jan 2026) — runs alongside the CLI for async background tasks. Beta for individual users; GA expected mid-2026.

Recommended

Use cases

Terminal-native developers running multi-step refactors from the shell
Backend engineers who want a CLI-first companion that respects git semantics
Government contractors needing FedRAMP-validated AI tooling
Non-coding knowledge work (research summaries, document drafts) via the same CLI
Mid-size teams piloting agentic workflows before broader rollout

Score caps

Risk flags

Reliability complaints
trustTemporary
Widespread reliability complaints (breaks often, unreliable output)
Caps Autonomy at 60
Removed when — 90+ days of improved reliability with community acknowledgment

Assessment

Status rationale

Validated because enterprise adoption criteria fully met: 70% Fortune 100, 8 of Fortune 10, named production deployments (NYSE, Stripe, Ramp, Netflix), comprehensive compliance (SOC 2 Type II, ISO 27001/42001, HIPAA BAA, FedRAMP High + DoD IL4/5 via Bedrock GovCloud, NIST CUI attestation). reliability-complaints cap is temporary and does not trigger demotion. Three command-injection CLI CVEs are vendor-dismissed as Informative with local/CI-precondition exploit paths and remain unpatched; monitored but not status-blocking.

Watch for

Movement triggers

Upgrade autonomy if: Opus 4.8 reliability gains hold through a 90-day production window with no new post-rollout degradation cycle AND community acknowledges improvement AND throttling/usage-limit backlash resolves (real-time weekly-cap visibility shipped). Upgrade compliance if: command-injection CLI CVEs (#35020/#35022) patched or formally mitigated AND FedRAMP on Claude Code CLI ships AND enterprise SLA added.

Downgrade to Assessed if: (a) Opus 4.8 triggers an April-scale community reliability crisis (2,000+ report equivalent, major press), OR (b) named enterprise customers publicly churn citing reliability, OR (c) a new critical unpatched CVE is actively exploited in the wild, OR (d) usage share falls below 50% primary-tool. Downgrade compliance/status if: CLI CVEs exploited in wild; Pentagon designation expands to commercial procurement.

Caution

Risks & limitations

Reliability Complaints
Moderate
Radar cap: reliability-complaints

Capabilities

Integration surface

Context window

200K tokens standard · 500K Enterprise · 1M beta (gated).

9 access points

Terminal CLIGA
Web interfaceGA
Desktop appGA
VS Code extensionGA
JetBrains pluginBeta
Slack integrationGA
GitHub ActionsGA
iOS mobileBeta
Cowork desktop agentBeta

Proof points

Adoption & benchmarks

80.9%

SWE-bench Verified

Highest in catalog

77.2%

Sonnet 4.5 (fallback)

Same benchmark

59.3%

Terminal-bench

Tooling agility

66.3%

OSWorld

Cross-OS task completion

Enterprise deployments

Accenture· 30,000 users
Cognizant· 350,000 deployment
TELUS· Enterprise rollout
Netflix· Engineering teams
Spotify· Engineering teams
Snowflake· Engineering teams

Government credentials

FedRAMP High, IL4/5, DoD selection, GSA OneGov pricing.

Productivity research

2–6 hours/week saved (Anthropic internal study, n=80). METR RCT (July 2025) showed 19% slowdown for experienced open-source developers in unfamiliar codebases — context-dependent.

Spotted something wrong or missing here? Suggest a change →