Mohammed Junaid

building with claude + ai agents, sharing what works

[Mohammed Junaid @mdjunaidap]

Claude built your app in 2 hours.

Give it 20 more minutes so it doesn't get hacked in 2 days.

Ask Claude to:

- Hide API keys
- Purge secrets from Git
- Expose only the public DB key
- Enable row-level security
- Encrypt sensitive data
- Enforce server-side auth
- Lock record access
- Block field tampering
- Secure session cookies
- Hash passwords
- Rate limit login
- Add bot protection
- Parameterize queries
- Validate all input
- Escape user content
- Restrict file uploads
- Trim API responses
- Add security headers
- Force HTTPS
- Scan dependencies

Vibe coding is fun.
Getting hacked isn't 🔒

[Mohammed Junaid @mdjunaidap]

@Hiteshdotcom the "to pass" line is such a good touch

half the frustration in labs is not knowing what the checker actually wants

[Mohammed Junaid @mdjunaidap]

@GoogleAIStudio "calibrated" is the key word here

knowing when the model is unsure matters more than the answer for most real workflows

[Mohammed Junaid @mdjunaidap]

@businessbarista step 5 is where it gets real

nobody budgets for the data cleanup, then it ends up being most of the cost

[Mohammed Junaid @mdjunaidap]

You can now mod Claude Code.

→ Change how it behaves
→ Customize the UI
→ Add your own features

Write one in a few lines of
TypeScript, or ask Claude to.
Mods ship inside plugins,
so you install them with /plugin.

3 mods I'd build first:

1. A token-spend status line
2. A guard that blocks .env edits
3. A "run tests before done" rule

What would you build?

[Mohammed Junaid @mdjunaidap]

The 6 agent design patterns
from Anthropic, in plain English:

1. Prompt chaining
step 1's output feeds step 2

2. Routing
route each input to its handler

3. Parallelization
run pieces at once, merge results

4. Orchestrator-workers
one LLM splits the job,
others do the work

5. Evaluator-optimizer
one writes, another grades, repeat

6. Autonomous agent
the model plans and acts in a loop

Most "agents" should be #1 or #2.

Reach for #6 last, not first.

Full article:
https://t.co/Fmo1a16eZs

[Mohammed Junaid @mdjunaidap]

@dannypostma seeing the ui update live while the agent works is the big one

remote is great for long tasks, but frontend tweaks feel faster locally

[Mohammed Junaid @mdjunaidap]

@michael_chomsky the replies are about to turn into a free launch board for every sandbox startup lol

[Mohammed Junaid @mdjunaidap]

@hwchase17 funny how it keeps coming back to the oldest skill

clearly saying what "done" looks like

[Mohammed Junaid @mdjunaidap]

More than half your users
will open your app on a phone.

Before launch, ask Claude to:
- Test every page at 375px wide
- Make tap targets 44px or bigger
- Add alt text to images
- Check color contrast
- Make it keyboard-friendly
- Show clear form errors
- Add loading states to buttons

"Looks good on my MacBook"
is not QA.

[Mohammed Junaid @mdjunaidap]

@Shpigford makes sense to start there tbh

"is it open, do you have it, how much" is exactly what people ask about a shop. software questions are way messier

[Mohammed Junaid @mdjunaidap]

@techie_piyush rough when your whole product depends on their api limits

hope someone from the dev relations team sees this

[Mohammed Junaid @mdjunaidap]

@coreyhainesco point 2 is the real work

most businesses can't explain their own pricing and faqs clearly. the ai will just repeat that mess back to customers

[Mohammed Junaid @mdjunaidap]

Claude isn't a chatbot anymore.

It's a whole workplace.

- Claude → think, write, plan
- Claude Code → build software
- Cowork → hand off real tasks
- Claude in Chrome → browse and act for you
- Claude in Excel → work inside your spreadsheets
- Whiteboards → map out messy plans
- Sheets → trackers and budgets that work
- Data → dashboards and data stories
- Artifacts → apps and pages you can share
- Connectors → Google, Adobe, Atlassian
- Skills and plugins → teach it how your team works
- Memory → it remembers how you like things done
- Fable, Mythos, Opus, Sonnet, Haiku
→ a model for every job

One subscription.
One assistant.
Your whole workday.

We went from "ask AI a question"
to "AI runs the work with you" 🚀

[Mohammed Junaid @mdjunaidap]

Microsoft spent 35 years teaching us:

Word for writing.
Excel for numbers.
PowerPoint for showing.

Claude just did it in one chat:

Whiteboards for thinking.
Sheets for tracking.
Data for showing.

You don't open 3 apps anymore.

You just say what you need.

[Mohammed Junaid @mdjunaidap]

Whiteboards and Sheets are rolling out in beta on paid plans.

Try this today:
Take the messiest plan in your head and say it out loud to Claude.

Then watch it become a board.

Pure magic ✨

[Mohammed Junaid @mdjunaidap]

Claude can now draw your thinking.

It's called Claude Whiteboards.

You talk through a messy plan.
Claude turns it into a clean board your whole team can edit.

Here's why this is bigger than it sounds 🧵

[Mohammed Junaid @mdjunaidap]

Before, AI gave you answers.

Now it gives you something to work on together.

That's the shift.

From chat
to workspace.