Sahan De Silva

Applied AI @ MacroActive

[Sahan De Silva @sahan_dsh]

If you use Claude Code across multiple sessions, do yourself a favour and add the active model to your status line.

I’ll switch one session to Fable, open another, completely tune out the startup header, and forget Fable is still selected then wonder where all my tokens went.

[Daniel Beauchamp @pushmatrix]

Everyone's talking about AI-generated HTML.

But have you tried giving your sites a zero-config API for saving data, file storage, AI, websockets, etc?

We did this at Shopify. Runs on a single VM that costs $200/month, and it's changed the way we work.

We call it Quick 👇🧵

[Sahan De Silva @sahan_dsh]

The first question I ask after an LLM gives me a plan before even reading it is

"Are there any items here being built way ahead for potential future use, but not needed for the current functionality?"

[Matt Pocock @mattpocockuk]

Everyone who's worked with AI a lot agrees: LLM's hallucinate. A LOT.

But to this day, I see friends and family trusting ChatGPT blindly, whether it's is working off retrieved data or not.

So, I made this. Share it with anyone who you think trusts AI too much.

[Google Gemini @GeminiApp]

Introducing Lyria 3, our new music generation model in Gemini that lets you turn any idea, photo, or video into a high-fidelity track with custom lyrics.

From funny jingles to lo-fi beats, you can create custom 30-second soundtracks for any moment.

See how it works. 🧵

[Cloudflare Developers @CloudflareDev]

In case you missed it, we made the new Workers best practices guide an agent skill.

Just run this to install it: npx skills add cloudflare/skills

[Sahan De Silva @sahan_dsh]

The new way I evaluate software: tell AI my exact requirements, ask for screenshots of each option, and get a feel for the UX without visiting a single website.

[Nikunj Kothari @nikunj]

https://t.co/3mnuYn7jfP

[ℏεsam @Hesamation]

"the people who struggle with coding agents are those who try to push their way too hard. partially why I find coding with agents easy is that I've led engineering teams before".

this is a unique view on AI coding, that to maximize your output you have to "let go" a bit and accept that the way AI will build the project might not be 100% aligned with your way, it might not even be 100% correct, but it will push the project forward.

and if you realize later that you don't like an approach, you can always go back to it later and iterate over it. so there's not always an absolute total gain, but the overall performance is what matters in the end.

[Sahan De Silva @sahan_dsh]

One tension I keep running into with agentic coding:

It’s great at shipping features.
It’s weak at protecting architecture.

I use coding agents daily. I let them write large parts of the system. It’s fast. It works. Then I go back and refactor.

Here’s the pattern.

You start with a simple function that queries data by date range. Later you need timezone awareness. The agent adds a new method. No breaking changes. Everything still works.

But now you have parallel logic.

In a short-lived script, who cares. In a product you’ll maintain for years, that’s technical drift.

A human engineer would usually refactor the original abstraction, update tests, and centralise the logic so future rule changes happen in one place. That’s not perfectionism. That’s long-term risk management.

So my current workflow is:
Let the agent move fast.
Then I consolidate.

But that’s manual oversight. And as projects grow, that becomes fragile.

If we’re serious about using agents for long-term systems that require ongoing upkeep, we need something beyond code generation. We need architectural judgement loops.

I’m actively trying to figure out what that looks like.

Are people building explicit refactor passes into their workflow?
Using architectural constraints in prompts?
Running periodic consolidation cycles?

Or are we quietly accepting structural sprawl as the trade-off for speed?

[Sahan De Silva @sahan_dsh]

This seems like the closest free alternative to WisprFlow I’ve seen: https://t.co/EcNUA2Z0uu

[Warren Sharp @SharpFootball]

“dinosaur documentary narrated by Morgan Freeman…
…produced by Steven Spielberg”

you had me at hello

[Lydia Hallie ✨ @lydiahallie]

Claude Code now supports agent teams (in research preview)

Instead of a single agent working through a task sequentially, a lead agent can delegate to multiple teammates that work in parallel to research, debug, and build while coordinating with each other.

Try it out today by enabling agent teams in your settings.json!

[Sahan De Silva @sahan_dsh]

Bash script - https://t.co/ZLF0IAKmby

[Sahan De Silva @sahan_dsh]

My terminal setup for coding with Claude Code & Zellij:

cd into project → run `zl` → get 3 tabs (Claude, Dev, Shell)

Session per project. No more tab chaos. (Bash script in the comments)

[Sahan De Silva @sahan_dsh]

Claude helped me to cancel 3 subscriptions this month.

- Make (Integromet): Exported the blueprints, had CC recreate the automation in code.

- CleanShot X Cloud: Built a small CLI to watch my desktop, upload to Cloudflare R2, share links, transcode videos.

- Notion: Rebuilt my PKM with proper RAG search on Convex.

[Sahan De Silva @sahan_dsh]

Depends on your output format Claude will generate something similar to below.

[Sahan De Silva @sahan_dsh]

Writing how-to guides used to be tedious. Then I found a workflow that allows Claude to handle it for me.

Simply annotate screenshots as you work, feed them to Claude, and receive a polished guide.

Here's how it works. 🧵

[Sahan De Silva @sahan_dsh]

Step 2: Ask AI to generate the guide

"The current folder contains a bunch of screenshots. The filenames indicate the order the steps should be executed. Generate a step-by-step guide explaining how to opt out of LinkedIn's AI data training based on these screenshots."