19. i build one ai thing a day. some of them work.
My life ➡️
Joined October 2017
- Tweets5.7K
- Following186
- Followers142
- Likes7K
Best part: editors using it daily shaped the features.
Hover to preview, scrub like the Project panel, cancel a download midway.
Even 2K/4K imports clean.
Talha retweeted
there are a lot of benchmarks that suggest 5.6 sol is the best model in the world right now, but the most reliable way to tell is that elon is obsessed with me again
Anthropic might just make Fable 5 stay for longer.
After the release of GPT-5.6-Sol, the market is on mixed opinions on both the models.
If Anthropic really cares about their users, they won’t let Fable go in API only.
Talha retweeted
Today we are launching Muse Spark 1.1, an upgrade to muse spark 1 that greatly improves agentic, coding, multimodal, and computer use capabilities. We're also launching the Meta Model API in public preview.
ai.meta.com/blog/introducing…
Spent today building a translation agent in copilot studio for a client.
Translated English product names to German, French, Italian, Spanish, and Danish; checked against their glossary first, ai fills gaps.
the ai part took an hour. power automate took the rest of the day:
- excel "tables" that weren't actually formatted as tables (silent failures)
- a column name with a space in it broke every filter query
- my first flow called the api 20 times for 5 languages before i replaced it with one flow returning all 5 at once
- a flow that refused to show up in copilot studio until i put it in a solution
Claude Cowork is insane.
I just made it do my 14 university assignments within an hour autonomously.
And the best part is, till 3rd August it’s on 2x usage.
What have you tried with Cowork yet?
i have a pipeline that ideates, renders, and publishes 2 short videos a day with zero input from me.
The scariest part wasn't building it; it was trusting it.
I spent almost a week developing this, so it produces a human-like feel
Here's the workflow I'm using:
3/4
Here is the place where the magic begins:
- Gemini watches those trending reels
- Gemini gets access to the background clips and audio clips (summarized)
- Gemini then creates a new task with it's best choice of font selection, content style, post idea and caption.
Today's build: the AI tutor now writes on the board in perfect sync with its own voice.
The problem: Gemini Live generates faster than real-time; voice, captions, and every drawing command arrive in one instant burst. The tutor was saying step 1 while step 5 was already on the board. the api gives you no timestamps to fix this.
The fix: use the audio itself as the clock. tag every caption and pen stroke with how much audio had arrived when it was generated, then release each one only when playback actually reaches that point.
Now it writes each equation exactly as it speaks it. Feels like a real teacher.
The trick behind yesterday's demo: the whiteboard is just tool calls.
the model gets draw_diagram, write_math, and clear_board as tools, and decides on its own mid-sentence when to use them. i never tell it when to draw; it just knows a graph helps here.
a custom layout engine keeps it from scribbling everything on top of itself.
