CG Nguyen

Sr Backend Eng. Building AI agents in Go. Shipped 3 MCP/Apify actors in May. Notes from the trenches.

[CG Nguyen @luongs3]

Senior backend eng, 10yr Go.

Last 9 days I shipped 3 AI-agent tools:

→ wg-gesucht-de — Berlin apartment scraper
→ germany-visa-jobs — DE visa-tech jobs pipeline
→ github-repo-intel-mcp — MCP for repo intel

Daily notes on what is working (and what is not) ↓

[CG Nguyen @luongs3]

#KiroweenHackathon26 I measured link rot on 2012 Hacker News front-page links: 55.2% are dead or moved (vs ~10% for a 2025 control). Then I resurrected 84.2% of them via the Wayback Machine.

Built with @kirodotdev (Resurrection category) 🎃

https://t.co/ZiC6rB4H1D

[CG Nguyen @luongs3]

I let @kirodotdev spec-drive a link-rot autopsy: 55% of 2012 Hacker News front-page links are dead or moved, and Wayback resurrects 84% of them. Best Kiro bit: the spec caught my own wrong metric twice before code shipped.

https://t.co/ZiC6rB4H1D
#KiroweenHackathon26

[CG Nguyen @luongs3]

Exploring how to build and test AI agents without input data, a significant challenge in the legal industry.

https://t.co/0Z7y0op69I

[CG Nguyen @luongs3]

Does NEAR Intents volume pull NEAR's price up? We tested 89 weeks of data.

Intents earned 48x NEAR L1 fees last quarter ($17.4M vs $0.36M), but volume growth doesn't lead BTC-adjusted NEAR price (r=+0.12, p=0.25).

https://t.co/63KwVSQIcg
#NEARHackathon @NEARProtocol

[CG Nguyen @luongs3]

Research shows that deterministic guardrails outperform smarter models. Hierarchies work better at scale, while multi-agent organizations often fail in human ways.

https://t.co/xLqET7AnoR

[CG Nguyen @luongs3]

agent-workforce-manager: a tool to deploy Codex and Claude agents for business goals like GTM and SEO. This can streamline your AI agent management.

https://t.co/KZ8ZVgZ8gV

[CG Nguyen @luongs3]

Andrew Ng is diving into AI Engineering, sharing insights that could shape the future of AI development and engineering practices.

https://t.co/CPZVSIFMnq

[CG Nguyen @luongs3]

@jackvi810 bro giờ làm gì để kiếm sống vậy?

[CG Nguyen @luongs3]

@SaharaAI @solana @HeySorinAI scammer!

[CG Nguyen @luongs3]

@SaharaAI stop talking shit and do nothing scammer

[CG Nguyen @luongs3]

rac-core enables agents to record and utilize decisions made by your team, enhancing consistency and effectiveness in agent responses.

https://t.co/DquZYgtGx6

[CG Nguyen @luongs3]

@Hivemapper Scam project. They didn't work for 1 month recently. Do not buy this token guys.

[CG Nguyen @luongs3]

I built an MCP so the agent does that part too. One tool call, your app is live on your own VPS with HTTPS: https://t.co/C22oUWv9An

[CG Nguyen @luongs3]

Vibe coding has a last-mile problem. The agent writes the whole app in 20 minutes, then you spend 3 hours on DNS, TLS, nginx and systemd to get it onto a real server. The bottleneck moved. It's not writing the code anymore. It's shipping it.

[CG Nguyen @luongs3]

Entry points. Where execution starts. An agent that knows main.go from a random util wastes far fewer steps getting oriented.

[CG Nguyen @luongs3]

Context engineering for coding agents, concretely. What actually belongs in the context window before the agent writes a single line:

[CG Nguyen @luongs3]

Hot files. The 5-10 files that matter for the task, not all 400. Token budget is real. Spend it on signal, not noise.

[CG Nguyen @luongs3]

I built an MCP that hands your agent that map automatically: https://t.co/z2AXz86W79

[CG Nguyen @luongs3]

Most AI coding agents fail for a boring reason: they start editing a repo they don't understand. No mental model of the structure or conventions, so they hallucinate paths and break things. The fix isn't a smarter model. It's giving the agent a map of the repo before it edits.