Just enjoy building and shipping AI SaaS 🚀 https://xitter.cf/t.co/Lh9SN94RR4
Egypt
Joined August 2026
- Tweets1.2K
- Following1.3K
- Followers776
- Likes757
Pinned Tweet
Your app can be useful.
But if the App Store screenshots don’t stop the scroll, almost no one downloads it.
Here’s the set we designed for Dosis.
I'm looking to connect with more people in tech & ai.
20, Solo founder from Egypt.
Building with Claude Code & Codex and sharing what I learn along the way.
If you're building too, say hi 👇
🚨 claude users !!
Stop running EVERYTHING on Opus !!
What Anthropic says each one is for:
Opus 5.5: hard agentic coding
Sonnet 5.5: 30%+ faster than Sonnet 5, cheaper
Haiku 5.5: subagents, summaries, browser use
Which one is your default?
A hiring manager applied to their own company’s job posting under a fake name.
Rejected in 6 minutes.
If you’ve been getting instant rejections from roles you’re clearly qualified for, it might not be you.
Just submitted an app to the App Store. Now it's sitting in "Waiting for Review".
How long did your last review take?
61% of essays by non-native English writers got flagged as AI.
Same detectors. Near-perfect on essays by native speakers.
If English is your second language, your score means something different.
Here's why, and what to do about it 👇
#AIDetection #ESL
Here's the line I keep coming back to:
An AI detector measures style. Not authorship.
97% of the TOEFL essays in that study got flagged by at least one of 7 detectors.
Every single one was written by a human student.
Cleanest demo of why short-text scores wobble.
A+B human. A+C AI. B+C AI.
Same sentences. Opposite verdicts.
Most detectors judge the whole window, and three sentences is barely a sample.
Treat a 100% on tiny text as noise, not proof.
A human wrote every word. The detector still flagged AI.
New research: people wrote stories from AI-made outlines.
A style detector caught 8%.
An idea detector caught 68%.
AI detection is splitting in two. Here's what changes 👇
#AIDetection #AIwriting
The line I'd tape to every review desk:
A detector score is a lead, not a verdict.
When AI wrote from a detailed human outline, the style detector flagged 92%. IdeaLens flagged 7%.
Both were "right." They measured different things.
Adding a remote MCP server to Gemini CLI takes 1 key in a settings file.
You can still lose an hour on it.
Not because it's hard.
Because of 2 traps nobody warns you about 👇
#MCP #GeminiCLI
TRAP 2: copied configs.
Gemini CLI spots a remote server by its "url" key. Add "transport": "sse" and it throws an unrecognized key error.
A config that works in one client tells you nothing about the next one.
❌ Don't paste. ✅ Use one url key, then /mcp auth.
Walter is a clean first MCP to test this on because it signs in with OAuth, so no API key sits in your settings file.
Run detect on a paragraph. A 0-100 score back means it works.
Gemini CLI setup steps: docs.walterwrites.ai/mcp-ser…
