Adi Dhar

following my curiosity. occasional writer, builder, ai @jpmorgan. prev @goldmansachs & @mckinsey 🇺🇸 🇮🇳. views my own

[Adi Dhar @Adi__Dhar]

10 Upgrades I Want in My AI Twin

I use AI daily to brainstorm, write, plan, build. But even with all the progress, they still fall short of what I’d call a true “AI twin.”

Today’s models are talented assistants, helpful, articulate, fast. But not yet adaptive. Not yet reflective of how I work.

Here are 10 things I wish my AI twin could already do 👇

1️⃣ Remember Me, Precisely 🧠
Not infinite memory, but modular memory. Let me choose what’s stored (tone, frameworks) and what’s not (medical history).
Let me spin up memory packs, carry them across models with a secure key.
Personalization should be composable.

2️⃣ Know My Digital Terrain 🌄
Don’t make me search for links. With permissions, crawl my Drive, Notion, Gmail. If I say “everything I’ve written on onboarding,” it should just pull the right docs. No handholding.

3️⃣ Connect Distant Dots ⚙️
Scoping a new idea? Pull in:
•A teardown on system inefficiencies
•An old Slack debate on workflows
•A buried memo from 6 months ago
Synthesize across time and format.

4️⃣ Pick the Right Model for the Task
Some models are better at convo, some at reasoning.
My twin should route smartly: “This prompt feels logic-heavy. Want o3?”

5️⃣ Fluidity Across Mediums
Drop in a doc → get a podcast.
Feed it Python → get a PPT critique.
Format should never be a blocker.

6️⃣ Ask Better Questions ❓
Sometimes I don’t need answers. I need a mirror.
“Are you solving for speed or avoiding the hard convo?”
Think Oura ring, but for thought patterns.

7️⃣ Push Back Where I Need It 🎯
Know where I’m decisive vs. where I stall.
Sharp at storytelling but verbose? Push me on pith.
Adapt without external prompting.

8️⃣ Act, Don’t Just Assist 🤖📞
If Gmail sees I’m 6 threads deep with a service rep, it should:
•Draft the reply
•Pick up my tone
•Send it (with permission)
•Chase a resolution

9️⃣ Train My Thinking, Too 🧩📈
Add “reasoning-lite” mode: I go first, it scaffolds. Track how my thinking evolves. Stretch me when I repeat blind spots. It’s resistance training for cognition.

🔟 Run As Me, With Rules 🧍⚡
Assign it workflows: investor updates, recruiting, scheduling.
It should follow my tone, my escalation logic, and what not to say.
I want to trust it to run that lane.

The next leap in AI adoption won’t just come from smarter models.
It’ll come from systems that adapt to us.

We’re close. But not there yet!

[Adi Dhar @Adi__Dhar]

@signulll Totally agree. Once token efficiency starts going back up, token limits will recede from the conversation, but this discount Walmart-esque branding will persist.

[Adi Dhar @Adi__Dhar]

@jenzhuscott I’d be interested in seeing the $ overlay here. How are the apps with significant usage doing on ARPU? If there’s a dent compared to pre-AI, we know the new app deluge has eaten away margins and surely sacrificed absolute traction along the way.

[Adi Dhar @Adi__Dhar]

Disagree. 3 pts: a) Most large cos' new code/total code share is tiny; token use restricted in numerator limits impact. b) Tech debt + platform sprawl erode any prod uplift via coord costs. c) Too early in cycle to measure AI-led gains—intangible/invisible vs ledger txns (for now).

[Adi Dhar @Adi__Dhar]

@PradyuPrasad The information assymetry is huge and it’s only going to get huger

[Adi Dhar @Adi__Dhar]

@0xchromium Gotta check this out pronto

[Adi Dhar @Adi__Dhar]

@boardyai Facts.

[Adi Dhar @Adi__Dhar]

@illscience Using a product obsessively is evergreen advice for folks of all ages. Good list!

[Adi Dhar @Adi__Dhar]

@jain_harshit There are whole Reddit threads discussing how to position yourself in the sweet spot of these leaderboards. TL;DR: Try to be in the middle of the top 10%.

[Adi Dhar @Adi__Dhar]

@sflorimm Org charts

[Adi Dhar @Adi__Dhar]

@tekbog Does it though? Using AI well to cause an impact requires some basic level of agency and will—traits often lacking in incompetent people.

[Adi Dhar @Adi__Dhar]

@signulll This is great idea. Apathy is an epidemic. Worse than pessimism

[Adi Dhar @Adi__Dhar]

@Noahpinion There’s a phenomenal @SemiAnalysis_ blog on this very topic today, quite on point.

[Adi Dhar @Adi__Dhar]

@ClaudeDevs This is bueno!

[Adi Dhar @Adi__Dhar]

Where does human judgement lie on the deterministic ↔ probabilistic spectrum? Is this a 1-D or multidimensional space? Any good papers on this?

[Adi Dhar @Adi__Dhar]

@shweta_ai I think that’s actually a trillion dollar idea.

[Adi Dhar @Adi__Dhar]

@dylan_curious Plus, the real question becomes: how do we measure AI's value beyond efficiency gains? The most interesting developments are happening in human-AI collaboration patterns.

[Adi Dhar @Adi__Dhar]

@dwarkesh_sp Couldn’t agree more. Plus, the citing is still top-notch.

[Adi Dhar @Adi__Dhar]

@devahaz That same $60 spent on groceries and a bit of home cooking easily yields 10–13 meals, that’s under $5 per meal. That’s the real comparison to a $60 restaurant meal.

[Adi Dhar @Adi__Dhar]

@signulll And Google’s no better. One random afternoon, Gmail hit me with an aggressive warning: “Pay more or stop getting emails in 24 hours”.

[Adi Dhar @Adi__Dhar]

@blakeaburge You just have more training data now!