The same pattern goes far beyond fitness.
CRM copilots.
Support agents.
Internal assistants.
Field operations.
Engineering tools.
Persistent business state + real-time environment can create a very different AI experience.
How are you handling context in your AI systems?
The biggest problem I’ve found with AI assistants isn’t the model.
It’s context.
How do you give an AI the right data, at the right moment, without rebuilding its context every time?
I’m testing this through my own fitness coaching system. 🧵
This is the part of Context Engineering I find most interesting:
It’s not about feeding an LLM more data.
It’s about deciding:
what should persist
what should be retrieved
what should be observed live
and when each context source matters
Example:
The machine planned for my exercise isn’t available.
Gemini already has the relevant fitness context from Drive.
I show Live the available machines and ask which alternative could fit the current workout.
That’s much more useful than starting from a blank chat.
Now there are two context layers:
Persistent context
→ program, workout history, previous sessions
Real-time multimodal context
→ the machines and equipment physically available around me
Same user. Same goal. Different context sources.
That’s where Gemini Live becomes useful.
Instead of relying on the Gem itself, I can have Live use the same fitness context stored in Drive.
Then I turn on the camera and show it the equipment available in the gym.
But the interesting part is what happens when my environment changes.
This week I’m training in a hotel gym.
Different equipment. Different machines.
My stored context knows my training history.
It doesn’t know what’s physically available around me.
That same context is used by a dedicated Gemini Gem for fitness coaching.
So when I ask about my training, I don’t need to re-explain:
my program
recent sessions
previous weights
workout notes
The context already exists.
My fitness app tracks workouts, exercises, sets, weights, reps, RIR and notes.
D1 remains the source of truth.
From that data, I generate a dedicated context document in Google Drive for the AI.
إذا AI Agent واحد يقدر يعوض جزء من خدمة engineer اليوم… شنو غادي يوقع ملي 5 أو 10 Agents يقدرو يخدمو مع بعضهم كفريق كامل، 24/7؟
واش كتشوفو هاد التطور فرصة ضخمة أكثر… ولا risk خاصنا ناخدوه بجدية من دابا؟
وهنا الفرق كبير جداً.
لأن Coding Agents وAI Agents ما بقاوش فقط chatbots كيعطيوك information.
كيقدرو يدخلوا فـworkflow، يكتبوا، يجربوا، يصلحو، يراجعو، ويعاودو يحاولو.
قبل عامين أو ثلاثة، السؤال الكبير كان: "شنو يقدر AI يجاوب؟"
دابا السؤال بدا كيتبدل لـ: "شنو يقدر AI ينفّذ ملي نعطيوه tools، codebase، simulations ووقت كافي؟"