Rakhul

Technical Owner (Applications) and AI Champion at Mercedes-Benz | Follow to learn how to use AI with Intent

[Rakhul @rakhul]

𝟵 Browser

Claude in Chrome.

Reading pages with you.

It clicks, navigates,
and extracts as you go.

Pair it with Cowork:

research to finished deliverable.
That's the full stack.

chat → reasoning → projects → skills
→ connectors → artifacts → cowork
→ code → browser

Nine tools. One weekend.

Saturday morning you're chatting.
Sunday night Claude is reading your files,
browsing the web, connected to your tools,
and running code in your terminal.

You'll never work the same way again.
Save this.

Share it with someone still using Claude
...like a search box.

If you want more playbooks like this,
...follow me @rakhul. I share these every week.

♻️ Repost if someone in your network
needs the full Claude map.

[Rakhul @rakhul]

I learned the full Claude ecosystem
in a weekend.

It took me weeks to figure out
..it was even possible.

Here's the guide I wish I had.
Claude isn't a chatbot anymore.

It's an entire ecosystem.
Chat. Code. Browse.

Automate. Build.
Most people use one corner of it.
This is the whole map.

Learnable in a single weekend.
9 tools. Saturday to Sunday.

[Rakhul @rakhul]

𝗨𝗡𝗗𝗔𝗬 𝗠𝗢𝗥𝗡𝗜𝗡𝗚

𝟳 Cowork

A desktop tool that reads
your actual files.
Excel. Word. PDF.

Multi-step tasks, handled.
Think of it as a digital ops assistant.

Point it at a folder. Give it a goal.

It runs.

𝟴 Claude Code

A command-line agent
that ships real files.

Reads your codebase.

Writes files. Runs commands.

Developers and technical founders,
....this is yours.

[Rakhul @rakhul]

THIS MODEL IS INSANE AT DESIGN. I ASKED GLM 5.2 (LEFT) AND OPUS 4.8 (RIGHT) TO BUILD ME A LANDING PAGE AND

GLM cost $0.06 while opus cost $0.49. More than 6x cheaper while being faster + more token efficient. Another win for open source AI.

[Rakhul @rakhul]

@aleabitoreddit The retrieval pollution problem is real and mostly ignored. RAG systems in particular have almost no signal for "is this source epistemically trustworthy" vs "is this text fluent and on-topic." Until that changes, recency + keyword match wins, and content farms know it

[Rakhul @rakhul]

@Polymarket If these docs hold up under review, the story isn't just Fauci, it's why NIH's grant oversight for dual-use research of concern was this porous for this long. EcoHealth Alliance was flagged internally in 2021. Nothing changed.

[Rakhul @rakhul]

@Vivek4real_ The number is interesting but it sidesteps the harder constraint: permitting, grid interconnection queues, and skilled labour bottlenecks will throttle gigafactory buildout long before capital does. Tesla's own Nevada timeline showed that.

[Rakhul @rakhul]

@BrianRoemmele @UnslothAI the unsloth quantisation pipeline deserves more credit than it gets. running full 70B class models at q4 on consumer hardware was a pipe dream 18 months ago. the real story isn't just "local AI", it's how fast the memory efficiency gap closed.

[Rakhul @rakhul]

@coinbureau The 60-day window was already fragile given how conditional Iran's domestic consensus was on the deal. Israeli action gives hardliners the exit ramp they wanted. Vance postponing Switzerland means this doesn't restart quietly, it restarts publicly or not at all.

[Rakhul @rakhul]

@unusual_whales That's the leverage play. ~21 million barrels a day pass through there. Even a "transit fee" of a few cents per barrel is worth billions annually, and it's almost impossible to enforce a refusal without military escalation.

[Rakhul @rakhul]

@Polymarket Boston doing this *for* the World Cup but those districts staying forever is the real story. Cities never roll back alcohol policy once the revenue is there.

[Rakhul @rakhul]

@rileybrown The "oh wait" ending is doing a lot of heavy lifting here. The real question is whether the agent marketplace model actually works, or if it just shifts the fragmentation problem from tools to agents.

[Rakhul @rakhul]

@RoundtableSpace The real question is what the compression does to semantic fidelity at the tails. 95% token reduction sounds incredible but if it quietly drops rare edge-case context, you won't notice until a bug slips through that the model would have caught with the full window

[Rakhul @rakhul]

@CryptooIndia The compliance angle here is underrated. HK's data residency rules and cross-border data transfer restrictions under PDPO make any US-hosted AI model a legal headache, not just a security one.

[Rakhul @rakhul]

@testingcatalog Persistent memory across sessions is the part worth watching. Most enterprise AI tools reset context daily, which is exactly why adoption stalls. The 3,000 tool integrations sound impressive until you ask how many orgs actually use more than 20 of them consistently.

[Rakhul @rakhul]

@BrianRoemmele The jump from "we make pretty images" to "we do full-body MRI" is massive. The regulatory path alone, FDA 510(k) or PMA, takes years. i'd want to see peer-reviewed sensitivity/specificity data before "routine as morning coffee" means anything

[Rakhul @rakhul]

@Vivek4real_ Gary Vee said the same thing in 2021 right before the 65% drawdown. the 13-year horizon is fine advice but it doesn't need a celebrity shouting it to be true, and it definitely doesn't need the caps lock