Corpus-Complete Convergence Analysis
Snapshot: corpus-complete-20260711 (2026-07-11) · source commit 1737631198b184e05ba0a343d67df6fa9e71b938 Scope: 1780 approved schema-v2 extractions; 1754 deduped sources; 991 deduped agentic projects.
Corpus-complete analysis output. Counts are generated from extraction.json; citations link to deterministic report.md views. This is an input to, not the output of, Phase 4 synthesis.
Convergence is measured at two levels. Building-block adoption rates answer whether projects converge on the same components; compressed solution-shape diversity answers whether they compose those components the same way.
Corpus-complete finding: production agentic systems converge on standardized bricks while remaining diverse in their custom buildings. The leading corpus-wide bricks are tool-use-orchestration 64.9%, eval-driven-development 47.9%, human-in-the-loop-gate 40.0%.
Denominator semantics: each problem class uses its deduplicated any-label membership. Classes overlap and are analyzed independently. Classes below 10 projects are labeled too small for a strong convergence read.
| Problem class | Deduped projects | Year trend | Top adopted blocks | Block-level read | Composition diversity | Dominant shape | Composition read |
|---|---|---|---|---|---|---|---|
| Evaluation, observability, and reliability governance | 1326 | 2021: 2 2022: 5 2023: 240 2024: 354 2025: 425 2026: 240 |
eval-driven-development 57.5%tool-use-orchestration 37.6%human-in-the-loop-gate 37.0%rag-retrieval 33.2% |
moderate component convergence | 74 shape(s); dominant 11.2% | multi-agent/tool workflow + evaluation-governed pipeline + guarded enterprise assistant | composition diverse |
| Enterprise knowledge retrieval and RAG | 1063 | 2022: 7 2023: 171 2024: 298 2025: 362 2026: 168 |
rag-retrieval 48.9%eval-driven-development 46.8%tool-use-orchestration 38.7%human-in-the-loop-gate 32.0% |
moderate component convergence | 74 shape(s); dominant 6.5% | retrieval-grounded assistant + multi-agent/tool workflow + evaluation-governed pipeline | composition diverse |
| Coding-agent and software delivery automation | 1022 | 2021: 1 2022: 2 2023: 168 2024: 236 2025: 323 2026: 239 |
eval-driven-development 55.6%tool-use-orchestration 47.0%human-in-the-loop-gate 39.1%observability-tracing 30.5% |
moderate component convergence | 79 shape(s); dominant 13.2% | multi-agent/tool workflow + evaluation-governed pipeline + guarded enterprise assistant | composition diverse |
| Infrastructure, cost, and latency optimization | 1010 | 2021: 2 2022: 8 2023: 158 2024: 248 2025: 339 2026: 204 |
eval-driven-development 46.8%tool-use-orchestration 37.0%human-in-the-loop-gate 30.4%observability-tracing 29.8% |
moderate component convergence | 74 shape(s); dominant 7.8% | multi-agent/tool workflow + evaluation-governed pipeline + guarded enterprise assistant | composition diverse |
| Context engineering and memory management | 754 | 2021: 1 2022: 2 2023: 97 2024: 172 2025: 249 2026: 192 |
tool-use-orchestration 52.0%eval-driven-development 50.1%human-in-the-loop-gate 35.7%rag-retrieval 33.2% |
moderate component convergence | 61 shape(s); dominant 11.9% | multi-agent/tool workflow + evaluation-governed pipeline + context/memory system | composition diverse |
| Safety, compliance, and trust controls | 688 | 2021: 2 2022: 2 2023: 112 2024: 201 2025: 247 2026: 89 |
eval-driven-development 46.7%human-in-the-loop-gate 44.3%rag-retrieval 39.0%tool-use-orchestration 38.7% |
moderate component convergence | 56 shape(s); dominant 13.7% | multi-agent/tool workflow + evaluation-governed pipeline + guarded enterprise assistant | composition diverse |
| Customer support and conversational service | 664 | 2022: 6 2023: 95 2024: 169 2025: 239 2026: 120 |
tool-use-orchestration 48.3%eval-driven-development 45.2%human-in-the-loop-gate 36.4%rag-retrieval 36.4% |
moderate component convergence | 66 shape(s); dominant 11.9% | multi-agent/tool workflow + evaluation-governed pipeline + guarded enterprise assistant | composition diverse |
| Document processing and structured extraction | 391 | 2021: 2 2022: 2 2023: 74 2024: 91 2025: 128 2026: 61 |
eval-driven-development 43.5%human-in-the-loop-gate 40.9%tool-use-orchestration 35.3%structured-output-contract 33.0% |
moderate component convergence | 68 shape(s); dominant 8.2% | multi-agent/tool workflow + evaluation-governed pipeline + guarded enterprise assistant | composition diverse |
| Agentic workflow automation and orchestration | 338 | 2023: 22 2024: 59 2025: 129 2026: 116 |
tool-use-orchestration 68.6%multi-agent-orchestration 47.9%human-in-the-loop-gate 44.1%eval-driven-development 40.8% |
strong component convergence | 41 shape(s); dominant 19.8% | multi-agent/tool workflow + evaluation-governed pipeline + guarded enterprise assistant | composition diverse |
| Domain-specific model adaptation | 67 | 2022: 2 2023: 6 2024: 21 2025: 19 2026: 14 |
fine-tuning-adaptation 85.1%eval-driven-development 47.8%cost-aware-routing 32.8%human-in-the-loop-gate 26.9% |
strong component convergence | 23 shape(s); dominant 16.4% | evaluation-governed pipeline + domain-adapted model pipeline + cost/latency optimized serving | composition diverse |
| Multimodal, edge, and voice AI systems | 67 | 2021: 1 2023: 15 2024: 15 2025: 22 2026: 10 |
fine-tuning-adaptation 31.3%model-routing 31.3%tool-use-orchestration 31.3%human-in-the-loop-gate 29.9% |
no dominant shared component yet | 36 shape(s); dominant 10.4% | multi-agent/tool workflow + guarded enterprise assistant + cost/latency optimized serving | composition diverse |
| Data analytics and decision intelligence | 57 | 2023: 5 2024: 19 2025: 22 2026: 7 |
tool-use-orchestration 59.6%human-in-the-loop-gate 35.1%rag-retrieval 33.3%eval-driven-development 26.3% |
moderate component convergence | 28 shape(s); dominant 12.3% | retrieval-grounded assistant + multi-agent/tool workflow + guarded enterprise assistant | composition diverse |
| Search, ranking, and recommendation quality | 50 | 2023: 4 2024: 14 2025: 21 2026: 5 |
eval-driven-development 54.0%rag-retrieval 32.0%cost-aware-routing 28.0%human-in-the-loop-gate 26.0% |
moderate component convergence | 29 shape(s); dominant 14.0% | retrieval-grounded assistant + evaluation-governed pipeline + cost/latency optimized serving | composition diverse |
| Content generation, localization, and creative operations | 40 | 2023: 10 2024: 13 2025: 11 2026: 5 |
human-in-the-loop-gate 60.0%eval-driven-development 50.0%structured-output-contract 27.5%tool-use-orchestration 27.5% |
moderate component convergence | 21 shape(s); dominant 12.5% | multi-agent/tool workflow + evaluation-governed pipeline + cost/latency optimized serving | composition diverse |
| Enterprise AI platform adoption and governance | 33 | 2023: 5 2024: 13 2025: 9 2026: 2 |
observability-tracing 51.5%tool-use-orchestration 45.5%rag-retrieval 42.4%privacy-preserving-processing 39.4% |
strong component convergence | 22 shape(s); dominant 21.2% | multi-agent/tool workflow + evaluation-governed pipeline + guarded enterprise assistant | composition diverse |
Strongest Two-Level Signals
Strong block convergence
- Domain-specific model adaptation: fine-tuning-adaptation 85.1%, eval-driven-development 47.8%, cost-aware-routing 32.8%; composition remains composition diverse.
- Agentic workflow automation and orchestration: tool-use-orchestration 68.6%, multi-agent-orchestration 47.9%, human-in-the-loop-gate 44.1%; composition remains composition diverse.
- Content generation, localization, and creative operations: human-in-the-loop-gate 60.0%, eval-driven-development 50.0%, structured-output-contract 27.5%; composition remains composition diverse.
- Data analytics and decision intelligence: tool-use-orchestration 59.6%, human-in-the-loop-gate 35.1%, rag-retrieval 33.3%; composition remains composition diverse.
- Evaluation, observability, and reliability governance: eval-driven-development 57.5%, tool-use-orchestration 37.6%, human-in-the-loop-gate 37.0%; composition remains composition diverse.
Highest composition diversity
- Enterprise knowledge retrieval and RAG: 74 shape(s), 6.5% dominant shape, top blocks rag-retrieval 48.9%, eval-driven-development 46.8%.
- Infrastructure, cost, and latency optimization: 74 shape(s), 7.8% dominant shape, top blocks eval-driven-development 46.8%, tool-use-orchestration 37.0%.
- Document processing and structured extraction: 68 shape(s), 8.2% dominant shape, top blocks eval-driven-development 43.5%, human-in-the-loop-gate 40.9%.
- Multimodal, edge, and voice AI systems: 36 shape(s), 10.4% dominant shape, top blocks fine-tuning-adaptation 31.3%, model-routing 31.3%.
- Evaluation, observability, and reliability governance: 74 shape(s), 11.2% dominant shape, top blocks eval-driven-development 57.5%, tool-use-orchestration 37.6%.
Limitations
- Shape signatures are deliberately compressed and now used only as a composition-diversity estimator.
- Block-adoption rates use normalized pattern vocabulary; unresolved proposed vocabulary terms are isolated in the normalization report.
- Year trends are metadata trends, not causal claims.
- Year trends are descriptive metadata trends; no causal interpretation is asserted.