Executive Snapshot
- 171 candidates scanned; 32 X + 20 YouTube + 25 Reddit + 30 dev web + 64 GitHub → enough to act; papers/Facebook partial → confidence medium.
- 2 OSS CLI repos dominate repo watch: opencode 166,442★, Codex 86,516★ → NEXA should benchmark OSS+commercial, not pick vendor by brand.
- TerminalBench/HN signal 393 pts / 148 comments around OSS agent performance → eval harness is now buying criterion, not research toy.
- 3 harness/spec signals in 24h HN sample (Superpowers, spec-as-source, VAEN) → SYNCA should treat prompts/specs/harness as versioned artifacts.
- 5 Fabbi domains impacted: FARE context, NEXA execution, SYNCA governance, DOMUS modernization, Japan/VN/Global delivery market.
KPI Dashboard
171
candidates
candidates
5/7
source groups usable
source groups usable
252,958★
opencode+codex
opencode+codex
5,291
Codex issues
Codex issues
72%
confidence medium
confidence medium
Executive Technical Signal
P0 Harness-first coding agents. Evidence: S03 VAEN 8/3, S06 TerminalBench 393/148. Counter: public benchmarks gameable. Fabbi: build internal benchmark from real Jira/PR tasks. Decision: trial. Next: 20-task suite.
P0 OSS CLI pressure. Evidence: S07 Codex 86,516★, S08 opencode 166,442★. Counter: issue backlog 11,540 combined. Fabbi: vendor-neutral NEXA adapter layer. Decision: adopt abstraction.
P1 Context/memory layer prerequisite. Evidence: S10 19,067★, S11 24,162★. Counter: stale/quality unknown. Fabbi: FARE should index repo+ADR+ticket history.
P1 Spec-as-source emerging. Evidence: S02, S04 59 pts/31 comments. Counter: low sample in last 24h. Fabbi: version prompt/spec/harness alongside code.
P2 Legacy API extraction. Evidence: S05 100 pts/83 comments. Counter: product-specific. DOMUS: test on 1 internal legacy workflow.
Trend Radar
| Zone | Signals | Move |
|---|---|---|
| Hot now | CLI agents, eval harness, spec workflows | Trial 2w |
| Emerging | portable harness packages, agent skills | Prototype |
| Watch | memory/task layers | Benchmark |
| Noise | generic agentic AI lists | Ignore |
| Declining | demo-only chat coding | Deprioritize |
KOL/OG Feed Watch
| Platform | Author | Time | Engagement | URL | Why CTO cares |
|---|---|---|---|---|---|
| HN | v-mdev | 2026-05-28T09:20Z | 1 pts / 0 comments | link | Skills/framework cho coding workflow → chuẩn hóa prompt/tooling như source artifact. |
| HN | OldDod | 2026-05-28T01:15Z | 1 pts / 0 comments | link | Spec-as-code tăng vai trò harness/spec review. |
| HN | sjhalani7 | 2026-05-27T20:52Z | 8 pts / 3 comments | link | Portable harness signal trực tiếp cho NEXA/SYNCA. |
| HN | cyrusradfar | 2026-04-01T18:32Z | 59 pts / 31 comments | link | Typed/FP guardrails được cộng đồng gắn với agent scaling. |
| HN | alexblackwell_ | 2026-04-16T15:19Z | 100 pts / 83 comments | link | Reverse engineering/API extraction liên quan DOMUS/legacy modernization. |
| HN | GodelNumbering | 2026-04-27T12:35Z | 393 pts / 148 comments | link | Benchmark-driven OSS agent cạnh tranh closed tool. |
| GitHub | openai | 2026-05-28T10:31Z | 86,516 stars / 12,661 forks / 5,291 issues | link | CLI coding agent adoption rất lớn; issue load cao → governance cần thiết. |
| GitHub | anomalyco | 2026-05-28T10:28Z | 166,442 stars / 19,811 forks / 6,249 issues | link | OSS coding CLI momentum mạnh; cần so sánh enterprise-readiness. |
| GitHub | bytedance | 2026-05-28T10:28Z | 69,850 stars / 9,424 forks / 924 issues | link | Multi-agent orchestration traction từ Asia builder. |
| GitHub | rohitg00 | 2026-05-28T10:31Z | 19,067 stars / 1,559 forks / 166 issues | link | Memory/context layer là prerequisite cho FARE. |
| GitHub | gastownhall | 2026-05-28T10:26Z | 24,162 stars / 1,613 forks / 392 issues | link | Task/context tracking cho AI coding workflow. |
| GitHub | getpaseo | 2026-05-28T10:25Z | 6,843 stars / 651 forks / 463 issues | link | Workflow/runtime repo có adoption nhưng issue risk. |
Repo Watch
| Repo | Metric | CTO judgment |
|---|---|---|
| openai/codex | 86,516 stars / 12,661 forks / 5,291 issues | CLI coding agent adoption rất lớn; issue load cao → governance cần thiết. |
| anomalyco/opencode | 166,442 stars / 19,811 forks / 6,249 issues | OSS coding CLI momentum mạnh; cần so sánh enterprise-readiness. |
| bytedance/deer-flow | 69,850 stars / 9,424 forks / 924 issues | Multi-agent orchestration traction từ Asia builder. |
| rohitg00/agentmemory | 19,067 stars / 1,559 forks / 166 issues | Memory/context layer là prerequisite cho FARE. |
| gastownhall/beads | 24,162 stars / 1,613 forks / 392 issues | Task/context tracking cho AI coding workflow. |
| getpaseo/paseo | 6,843 stars / 651 forks / 463 issues | Workflow/runtime repo có adoption nhưng issue risk. |
| mixpeek/amux | 205 stars / 25 forks / 0 issues | Nhỏ nhưng sạch issue; watch cho memory/tool routing. |
| DjangoPeng/agentic-ai | 89 stars / 68 forks / 0 issues | Education/reference signal; không đủ để adopt. |
Paper / Benchmark Watch
- arXiv collector: 0 usable items; reason: HTTP 429 + timeout; confidence impact: papers layer -15%.
- Benchmark proxy: TerminalBench discussion/repo signal 393 pts / 148 comments via HN/GitHub.
- Action: re-run paper collector with Semantic Scholar/OpenReview fallback next cron.
Product / Business Watch
| Product | Signal | Fabbi move |
|---|---|---|
| Claude Code/Codex/Cursor | N/A fresh product changelog in harness; GitHub Codex 86,516★ | Route through adapter, not direct dependency |
| OpenCode/OSS agents | opencode 166,442★; dirac HN 393/148 | Benchmark with internal tasks |
| Sourcegraph/Cody/JetBrains/Replit/Gemini | N/A product collector partial | Watch, no procurement action today |
Impact Coverage
| Domain | Now 0-2w | Next 1-2m | Later 3-6m |
|---|---|---|---|
| FARE | Index repo/ADR/ticket context | Memory eval set | Context quality SLA |
| NEXA | CLI adapter + 20 tasks | Model/tool routing | Agent execution platform |
| SYNCA | Prompt/spec/harness versioning | Risk gate in PR | Audit dashboard |
| DOMUS | Pick 1 legacy API extraction case | Human-in-loop modernization | JP legacy migration package |
| Japan/VN/Global | Position delivery ROI 15-25% | Benchmark proof for presales | Managed AI-SDLC offering |
CTO Recommendations
1. Build 20-task internal harness
ROI/time-saving: 18-25%; risk 2/5; owner: AI Platform Lead; TTV: 10 days; validation: pass@1, rollback rate, review minutes.
ROI/time-saving: 18-25%; risk 2/5; owner: AI Platform Lead; TTV: 10 days; validation: pass@1, rollback rate, review minutes.
2. Create vendor-neutral CLI adapter
ROI: 12-20%; risk 3/5; owner: NEXA Tech Lead; TTV: 2 weeks; validation: same task on Codex/opencode/Claude.
ROI: 12-20%; risk 3/5; owner: NEXA Tech Lead; TTV: 2 weeks; validation: same task on Codex/opencode/Claude.
3. Version prompts/specs/harness with code
ROI: 10-15%; risk 2/5; owner: SYNCA QA Lead; TTV: 1 week; validation: defect leakage delta.
ROI: 10-15%; risk 2/5; owner: SYNCA QA Lead; TTV: 1 week; validation: defect leakage delta.
4. Pilot FARE context memory
ROI: 15-22%; risk 3/5; owner: FARE Architect; TTV: 3 weeks; validation: context hit-rate ≥70%.
ROI: 15-22%; risk 3/5; owner: FARE Architect; TTV: 3 weeks; validation: context hit-rate ≥70%.
Action Plan
DO THIS WEEK
- 20 real tickets → eval harness.
- Compare 3 CLIs on same repo.
- Define 5 PR risk gates.
WATCH 2–4 WEEKS
- TerminalBench leaderboard movement.
- Codex/opencode issue burn-down.
- Semantic Scholar papers fallback.
IGNORE / LOW SIGNAL
- Generic “agentic AI” list repos.
- Demo videos without code/eval metrics.
Trend Momentum
CLI agents: ↑ 252k★; Harness: ↑ 3 fresh HN; Papers: ↓ 0 usable due 429.
CLI agents: ↑ 252k★; Harness: ↑ 3 fresh HN; Papers: ↓ 0 usable due 429.
Detailed Source Appendix
| ID | Platform | Title | Author | Time | Metric | CTO note |
|---|---|---|---|---|---|---|
| S01 | HN | Superpowers: An Agentic Skills Framework for AI Coding Workflows | v-mdev | 2026-05-28T09:20Z | 1 pts / 0 comments | Skills/framework cho coding workflow → chuẩn hóa prompt/tooling như source artifact. |
| S02 | HN | With coding agents, specs feel more like source code | OldDod | 2026-05-28T01:15Z | 1 pts / 0 comments | Spec-as-code tăng vai trò harness/spec review. |
| S03 | HN | Show HN: VAEN – Package and import portable AI coding-agent Harnesses | sjhalani7 | 2026-05-27T20:52Z | 8 pts / 3 comments | Portable harness signal trực tiếp cho NEXA/SYNCA. |
| S04 | HN | Functional programming accelerates agentic feature development | cyrusradfar | 2026-04-01T18:32Z | 59 pts / 31 comments | Typed/FP guardrails được cộng đồng gắn với agent scaling. |
| S05 | HN | Launch HN: Kampala – Reverse-Engineer Apps into APIs | alexblackwell_ | 2026-04-16T15:19Z | 100 pts / 83 comments | Reverse engineering/API extraction liên quan DOMUS/legacy modernization. |
| S06 | HN | Show HN: OSS Agent topped TerminalBench on Gemini-3-flash-preview | GodelNumbering | 2026-04-27T12:35Z | 393 pts / 148 comments | Benchmark-driven OSS agent cạnh tranh closed tool. |
| S07 | GitHub | openai/codex | openai | 2026-05-28T10:31Z | 86,516 stars / 12,661 forks / 5,291 issues | CLI coding agent adoption rất lớn; issue load cao → governance cần thiết. |
| S08 | GitHub | anomalyco/opencode | anomalyco | 2026-05-28T10:28Z | 166,442 stars / 19,811 forks / 6,249 issues | OSS coding CLI momentum mạnh; cần so sánh enterprise-readiness. |
| S09 | GitHub | bytedance/deer-flow | bytedance | 2026-05-28T10:28Z | 69,850 stars / 9,424 forks / 924 issues | Multi-agent orchestration traction từ Asia builder. |
| S10 | GitHub | rohitg00/agentmemory | rohitg00 | 2026-05-28T10:31Z | 19,067 stars / 1,559 forks / 166 issues | Memory/context layer là prerequisite cho FARE. |
| S11 | GitHub | gastownhall/beads | gastownhall | 2026-05-28T10:26Z | 24,162 stars / 1,613 forks / 392 issues | Task/context tracking cho AI coding workflow. |
| S12 | GitHub | getpaseo/paseo | getpaseo | 2026-05-28T10:25Z | 6,843 stars / 651 forks / 463 issues | Workflow/runtime repo có adoption nhưng issue risk. |
| S13 | GitHub | mixpeek/amux | mixpeek | 2026-05-28T10:24Z | 205 stars / 25 forks / 0 issues | Nhỏ nhưng sạch issue; watch cho memory/tool routing. |
| S14 | GitHub | DjangoPeng/agentic-ai | DjangoPeng | 2026-05-28T10:27Z | 89 stars / 68 forks / 0 issues | Education/reference signal; không đủ để adopt. |
| S15 | Data | Harness validation run | local harness | 2026-05-28T10:35Z | 171 candidates / 5 platform groups | Volume đủ; arXiv/Facebook partial giảm confidence papers/social-public. |
Data Quality / Scan Health Appendix
Scan: 171 candidates. Breakdown: X 32 search fallback, YouTube 20, Reddit 25, dev_web/HN 30, GitHub 64, papers/product 0 due arXiv 429/timeouts, Facebook public 0 no usable links. Gate: QUALITY_GATE_PARTIAL; reason: papers/Facebook unavailable, not enough to block because social+GitHub+HN volume ≥100.