athahar

Product Manager | x - Square, PayPal, Apple | co founded @shipwavesonline. (IPO’d Dec 2025). also building:

[athahar @athahar]

Right now people are manually checking dozens of destinations and flight statuses on FlightRadar, then repeating airline site searches one-by-one only to find many options are canceled or unavailable.

Built a live DXB Flight Availability tracker that filters scheduled/estimated departures, excludes Middle East destination/transit routes, and verifies ticket availability with direct airline booking links.

https://t.co/SAkq5lnojv

If this can help someone, please share and let me know your feedback so I can improve it quickly.

#DXB #FlightTracker #EmergencyTravel #DubaiAirport #dubai #Aviation #BuildInPublic

@MarioNawfal

[athahar @athahar]

@nikunj This is awesome, I would love to test for many destinations if you have an alpha access? curious which API are you using to translating the flights/hotel prices to points?

[athahar @athahar]

Radar feedback:
1. Radar - a txn with a risk score "normal" - got hit with a dispute. [also, smart dispute didn't recommend a refund, a la smart refund feature]
2. Radar - blocked a bunch of txns from a scammer with rotating IPs. Scammer changed the email, used one of the same IPs that was previously blocked, with a new card - Stripe approved txn (all with in 10mins).
3. Also, i feel i need stripe radar assistant to expand it capability to brainstorm the issues that i'm going thro and then help me recommend a rule that can help me, based on the list of suspect txns that i want to share with ( i do this with claude outside, but i feel Stripe can do better job with the business context, as well as payment / txn data across the universe)
happy to share more details with team if needed.

[athahar @athahar]

@jeff_weinstein i use https://t.co/2gJ5e8DqgX

[athahar @athahar]

Every UI ever built was a translation layer. The human mind had to transform a messy thought ("I want to see how my team did last week") into structured clicks: navigate → filter → date range → export.

That forced translation built a cognitive muscle.

When the chat bar removes that translation step, two things happen:

1. Users get what they want faster
2. That cognitive muscle quietly atrophies

We're celebrating #1. Nobody's talking about #2.

[athahar @athahar]

@petergyang maybe then Google Gemma 4?

[athahar @athahar]

@devneeddev Can you use Jannal for any additional overage call and we can see how much is the actual cost and tokens used?

https://t.co/AwSeCMGKUL



[athahar @athahar]

@OldBearLearnsAI @Google @grok @grok good for openclaw?

[athahar @athahar]

Many folks had to upgrade in the last few weeks. I had to upgrade to 200$ plan; $100 plan worked perfectly fine for all this while before the Anthopic’s Spring deal(that doubled usage limits outside non peak hours)

Can you ship a small semantic llm in Claude code which stays in client side which decide which models to use for the task instead of putting the onus on the user. That’s not their primary job to figure which models to use when when coding- some may do that for their llm API usage.

It’s like using automated gears vs stick gear in cars, where users had to decide which gear I should run on - that’s doesn’t scale and not a good UX.

@bcherny @trq212 @_catwu

[athahar @athahar]

Many folks had to upgrade. I had to upgrade to 200$ plan, all the while $100 plan has been good enough for all this while.
Can you ship a small semantic llm in Claude code which stays in client side which decide which models to use for the task instead of putting the onus on the user. That’s not their primary job to figure which models to use. It’s like using automated cars vs gear stick models in cars where users had to decide which gear I should run on when - that’s doesn’t scale and not a good UX.

[athahar @athahar]

@burkov Doesn’t it cache?

[athahar @athahar]

@jacob_posel @bcherny @claudeai See what’s going on in every request. Maybe a lot of MCP and tool calls

https://t.co/AwSeCMGKUL

[athahar @athahar]

@jasonzhou1993 @aibuilderclub_ Great thread. Also see where your tokens get used up in every request. (Inspect your Claude request like chrome tools inspector)

https://t.co/AwSeCMGKUL

https://t.co/vH4RQOFI2z

[athahar @athahar]

Most people think they need a higher Claude plan, when their usage runs out in 2 hours. They don't. They need fewer MCP tools loaded.

I have watched ~50% of my context window go to MCP  tool definitions that were never used, every time. (sometimes even higher)

Your context window is a budget. Every token spent on tool schemas is a token that can't be spent on your code, your conversation, your reasoning.

We built Jannal to make this visible and fixable. Give it a try and share your feedback.

[athahar @athahar]

@trq212 can you tell whats the token limits during that 5 hour session for each plan, so folks know it? i don't think just progress bar is not transparent.

[athahar @athahar]

Claude Code - API Error: Unable to connect to API (ConnectionRefused)

very consistent error today. any idea ?

@AnthropicAI @bcherny @trq212 @_catwu

[athahar @athahar]

@lennysan @Railway @resend @fastmcp 💯I would also add @posthog, @supabase and @expo to that list!!

[athahar @athahar]

@claudeai Thank you. I know it took long, but truly needed! Thank you @trq212 @bcherny

[athahar @athahar]

@Al_Grigor See what's eating your context window/usage.

Watch every API request in real time. See exactly how many tokens go to the system prompt, tool definitions, conversation history, and tool results

https://t.co/AwSeCMGKUL

[athahar @athahar]

@code_kartik @AnthropicAI See what's eating your context window/usage.

Watch every API request in real time. See exactly how many tokens go to the system prompt, tool definitions, conversation history, and tool results

https://t.co/AwSeCMGKUL

[athahar @athahar]

Do you know what’s eating up the tokens for every call. Jannal is the chrome dev inspector.

2 possibilities. A bunch of MCP tools are loaded for every call/context. Unnecessary.
That agent is calling a bunch of sub agents that eats up the tokens.
Try it and let me know. I’m one of the maintainers for this.

[athahar @athahar]

@BradGroux @AnthropicAI You can use this to check where and how the usage is happening. Just do npx install
Open source, local proxy

https://t.co/AwSeCMGKUL

https://t.co/vH4RQOFI2z

[athahar @athahar]

Most people think they need a higher Claude plan, when their usage runs out in 2 hours. They don't. They need fewer MCP tools loaded.

I have watched ~50% of my context window go to MCP  tool definitions that were never used, every time. (sometimes even higher)

Your context window is a budget. Every token spent on tool schemas is a token that can't be spent on your code, your conversation, your reasoning.

We built Jannal to make this visible and fixable. Give it a try and share your feedback.