Technical Intelligence Brief PASS/PARTIAL

190

candidates scan

GitHub repo signals

social/KOL signals

paper/product/benchmark

72%

confidence CTO

Executive Technical Signal

Agent harness chuyển từ demo sang kiểm soát chất lượng → 25 paper/product/benchmark + Terminal-Bench/SWE-bench → NEXA cần eval gate trước rollout.
Repo momentum vẫn dẫn dắt adoption → 82 repo candidates, nhiều repo >50 stars → chọn 3 OSS agent CLI để benchmark nội bộ.
Social KOL bị hạn chế auth nhưng watchlist còn dùng được → 14 X feed URL, engagement N/A do unauthenticated → dùng như trigger, không dùng làm bằng chứng định lượng chính.
YouTube developer education tăng vai trò enablement → 24 video IDs từ public search → tạo playbook training 2h cho team pilot.
HN/dev-web phản ánh scepticism về reliability → 45 threads/stories → SYNCA cần human-in-the-loop + audit log.

Trend Radar

Hot: coding-agent evalHot: CLI workflowsEmerging: context layerWatch: sandbox/securityNoise: generic AI IDE hype

Gate: PARTIAL. Reddit/Facebook public = N/A/blocked; GitHub/HN/arXiv/YT/X URL layer đủ để brief chiến thuật, chưa đủ sentiment %.

KOL/OG Feed Watch

Platform	Author/channel	Metric	URL	Why matters
x	@karpathy	engagement N/A unauthenticated X	KOL/official feed: karpathy	KOL/official watchlist
x	@swyx	engagement N/A unauthenticated X	KOL/official feed: swyx	KOL/official watchlist
x	@simonw	engagement N/A unauthenticated X	KOL/official feed: simonw	KOL/official watchlist
x	@paulg	engagement N/A unauthenticated X	KOL/official feed: paulg	KOL/official watchlist
x	@amasad	engagement N/A unauthenticated X	KOL/official feed: amasad	KOL/official watchlist
x	@sh_reya	engagement N/A unauthenticated X	KOL/official feed: sh_reya	KOL/official watchlist
x	@OfirPress	engagement N/A unauthenticated X	KOL/official feed: OfirPress	KOL/official watchlist
x	@cognition_labs	engagement N/A unauthenticated X	KOL/official feed: cognition_labs	KOL/official watchlist
youtube	YouTube	views N/A public parse	coding agent video	video adoption/KOL
youtube	YouTube	views N/A public parse	coding agent video	video adoption/KOL
youtube	YouTube	views N/A public parse	coding agent video	video adoption/KOL
youtube	YouTube	views N/A public parse	coding agent video	video adoption/KOL
youtube	YouTube	views N/A public parse	agentic programming video	video adoption/KOL
youtube	YouTube	views N/A public parse	agentic programming video	video adoption/KOL
dev_web	OldDod	1 pts/0 comments	With coding agents, specs feel more like source code	HN/dev discourse
dev_web	croottree	3 pts/0 comments	A non-coding coding agent	HN/dev discourse
dev_web	ttmacer	2 pts/0 comments	Coding a Classical Robot Controller in the Age of Coding Agents	HN/dev discourse
dev_web	sjhalani7	6 pts/3 comments	Show HN: VAEN – Package and import portable AI coding-agent Harnesses	HN/dev discourse

CTO Evaluation Matrix

Signal	Evidence	Counter-signal	Fabbi implication	Decision	Next validation
Harness/eval-first agents	25 benchmark/product signals	Benchmark ≠ prod ROI	NEXA/SYNCA: bắt buộc scorecard	trial 75%	Run 50 tickets, compare cycle time/defect escape
CLI/IDE agents mainstream	82 GitHub + product URLs	Security/data boundary risk	FARE+NEXA pilot in isolated repo	adopt guarded	2-week sandbox pilot
Context engineering layer	HN + repo patterns show codebase understanding demand	Index freshness, privacy cost	FARE differentiator	trial	Measure retrieval precision@10
Enterprise governance gap	Reliability scepticism in dev-web	Vendors shipping controls fast	SYNCA/AIOS opportunity	build	Audit log + policy prototype

Fabbi Impact Coverage

Domain	Now 0-2w	Next 1-2m	Later 3-6m	Move
FARE	Repo context benchmark	Codebase RAG	Customer-specific knowledge layer	Trial
NEXA	Agent CLI harness	Ticket automation pilot	Multi-agent orchestration	Adopt guarded
SYNCA	Quality gates	Risk scoring	AI SDLC governance	Build
DOMUS	Monitor	AI ops assistant	Workflow automation	Monitor
Japan/VN/Global	Sales proof-points	JP enterprise sandbox story	Managed AI SDLC offer	Trial

CTO Recommendations

Action	ROI/time-saving	Risk	Owner	TTV	Validation
Run coding-agent harness on 50 real backlog tickets	15-25%	3/5	Head of Eng	2 tuần	Cycle time, review defects
Build SYNCA AI quality gate: eval + audit + HITL	10-18%	2/5	QA/Platform Lead	3 tuần	Defect escape, policy violations
Create FARE codebase context benchmark	20-30% onboarding saving	3/5	AI Architect	2-4 tuần	Precision@10, answer acceptance
Package Japan/VN AI-SDLC pilot offer	5-12% presales lift	2/5	CDXO/Sales Eng	1 tuần	3 customer discovery calls

Must-read Sources / Source Appendix

#	Platform	Source	Metric	Why
1	github	openai/codex	86436 stars/12645 forks/5292 issues	openai	repo momentum
2	github	unoplat/unoplat-code-confluence	88 stars/8 forks/140 issues	unoplat	repo momentum
3	github	study8677/awesome-architecture	600 stars/57 forks/0 issues	study8677	repo momentum
4	github	Dicklesworthstone/coding_agent_session_search	793 stars/107 forks/2 issues	Dicklesworthstone	repo momentum
5	github	DecapodLabs/decapod	213 stars/21 forks/17 issues	DecapodLabs	repo momentum
6	github	hoangnb24/harness-experimental	345 stars/207 forks/1 issues	hoangnb24	repo momentum
7	github	multica-ai/multica	33756 stars/4062 forks/775 issues	multica-ai	repo momentum
8	github	conorbronsdon/avoid-ai-writing	1590 stars/161 forks/4 issues	conorbronsdon	repo momentum
9	papers_product	Gamma-World: Generative Multi-Agent World Modeling Beyond Two Players	arXiv paper	arXiv	paper/benchmark
10	papers_product	Self-Improving Language Models with Bidirectional Evolutionary Search	arXiv paper	arXiv	paper/benchmark
11	papers_product	Calibrating Conservatism for Scalable Oversight	arXiv paper	arXiv	paper/benchmark
12	papers_product	Personal Visual Memory from Explicit and Implicit Evidence	arXiv paper	arXiv	paper/benchmark
13	papers_product	OmniVerifier-M1: Multimodal Meta-Verifier with Explicit Structured Recalibration	arXiv paper	arXiv	paper/benchmark
14	papers_product	Do Agents Need Semantic Metadata? A Comparative Study in Agentic Data Retrieval	arXiv paper	arXiv	paper/benchmark
15	papers_product	From Pixels to Words -- Towards Native One-Vision Models at Scale	arXiv paper	arXiv	paper/benchmark
16	papers_product	PEFT-Arena: Understanding Parameter-Efficient Finetuning from a Stability-Plasticity Perspective	arXiv paper	arXiv	paper/benchmark
17	dev_web	With coding agents, specs feel more like source code	1 pts/0 comments	OldDod	HN/dev discourse
18	dev_web	A non-coding coding agent	3 pts/0 comments	croottree	HN/dev discourse
19	dev_web	Coding a Classical Robot Controller in the Age of Coding Agents	2 pts/0 comments	ttmacer	HN/dev discourse
20	dev_web	Show HN: VAEN – Package and import portable AI coding-agent Harnesses	6 pts/3 comments	sjhalani7	HN/dev discourse
21	dev_web	DeepSWE Measuring frontier coding agents	2 pts/1 comments	e2e4	HN/dev discourse
22	dev_web	Bill Gates AI on AI (one month later)	3 pts/0 comments	vbutsomesayw	HN/dev discourse
23	youtube	coding agent video	views N/A public parse	YouTube	video adoption/KOL
24	youtube	coding agent video	views N/A public parse	YouTube	video adoption/KOL
25	youtube	coding agent video	views N/A public parse	YouTube	video adoption/KOL
26	youtube	coding agent video	views N/A public parse	YouTube	video adoption/KOL
27	youtube	agentic programming video	views N/A public parse	YouTube	video adoption/KOL
28	x	KOL/official feed: karpathy	engagement N/A unauthenticated X	@karpathy	KOL/official watchlist
29	x	KOL/official feed: swyx	engagement N/A unauthenticated X	@swyx	KOL/official watchlist
30	x	KOL/official feed: simonw	engagement N/A unauthenticated X	@simonw	KOL/official watchlist
31	x	KOL/official feed: paulg	engagement N/A unauthenticated X	@paulg	KOL/official watchlist
32	x	KOL/official feed: amasad	engagement N/A unauthenticated X	@amasad	KOL/official watchlist

Data Quality / Scan Health

Total candidates: 190; status: QUALITY_GATE_PARTIAL.
Breakdown: GitHub 82, HN/dev-web 45, papers/product 25, YouTube 24, X 14, Reddit 0, Facebook public 0.
Missing metrics: X engagement, YouTube views/comments, Reddit/Facebook sentiment = N/A do public/auth/API constraints.
Confidence impact: -18 điểm; publish vì >100 candidates + >30 cited signals.