Databases are great, if you know what you are doing. AI agents will default to databases because it's what's in their training data, but I would argue that you should prompt it to use plaintext for data storage. I think plaintext is THE storage format for AI vibe-coded personal software. Here is my argument: https://shinglyu.com/blog/2026/04/25/plaintext-as-the-new-standard-for-personal-software.html #AI #vibecoding
@shinglyu
- 6 Followers
- 13 Following
- 95 Posts
In my latest post, I break down the configuration and the specific pitfalls I encountered—from firewall allowlists to the nuances of the runs-on label and permission requirements.
Check it out here: https://shinglyu.com/blog/2026/04/20/copilot-cloud-agent-custom-runner-aws-codebuild.html
Want to run GitHub Copilot Cloud Agents entirely within your own AWS VPC?
I’ve been experimenting with using AWS CodeBuild as a self-hosted runner for Copilot, allowing for a fully remote "vibe coding" setup. It's a great way to handle issue implementation in the background, but the setup has a few "sharp corners" you need to watch out for.
If you aren't looking into GitHub Agentic Workflows yet, you’re already behind the curve on the next evolution of CI/CD.
Read the full breakdown here: https://shinglyu.com/blog/2026/04/15/automating-weekly-research-with-github-agentic-workflows.html
#GitHub #GitHubAgenticWorkflows #AI #Automation #SoftwareArchitecture #DevOps
Instead of jumping between different AI tools and UIs, I’ve moved my entire weekly tech deep-dive into code and GitHub Actions. It’s automated, it’s headless, and it lives exactly where my work does.
In my latest post, I break down:
- Why I’m ditching UI-based AI tools for a code-first approach.
- How to orchestrate deep research directly within your repo.
- My exact setup for tracking AWS, AI trends, and Software Architecture updates on autopilot.
Most people are still trying to automate their lives through clunky web dashboards and "no-code" interfaces. But as developers, we know that if it isn’t in code and version-controlled, it’s a nightmare to maintain.
I’ve spent the last few weeks diving into GitHub Agentic Workflows, and it’s a total game-changer for how we handle agentic automation.
You get automation without blindly trusting a model with your numbers. Write the import rules once, and everything stays deterministic and auditable after that.
If you want more control over their financial data without paying for yet another SaaS tool, this might be the workflow you've been looking for.
Full post: https://shinglyu.com/blog/2026/03/18/hledger-and-ai.html
📊 Basic transactions & balance sheets — your full financial picture in a text file you actually own and version controlled
📈 Investment tracking with live market prices — query your portfolio value at any point in time
🇳🇱 Dutch Box 3 tax reporting — including foreign currency accounts (yes, even that TWD savings account)
The key insight that changed how I think about AI + finance: don't ask AI for the answer — ask it to write the hledger commands, then run them yourself.
📈 Investment tracking with live market prices — query your portfolio value at any point in time
🇳🇱 Dutch Box 3 tax reporting — including foreign currency accounts (yes, even that TWD savings account)
The key insight that changed how I think about AI + finance: don't ask AI for the answer — ask it to write the hledger commands, then run them yourself.
I am usually quite proud of my personal finance setup, quarterly balance sheets and monthly income statements in Excel with nice visuals. Until I found missing information and wrong calculations every tax season...
Then I switched to hledger — a plain text accounting tool — and started using AI to help set it up. Now my budgeting and tax reporting basically run themselves.
My new post covers 3 practical examples:
Do you write architecture decision records that just sit on the shelf and nobody follows it? Having them version controlled with the code and use AI agents to check every pull request might be a solution. See my experiment on using AI agent to review PRs based on ADRs. https://shinglyu.com/blog/2026/03/01/ai-adr-code-review.html
