Keyur Shah

Founder KramaAI | Workflow intelligence | Prev Product @Opendoor @Eventbrite | |

[Keyur Shah @keyur19198]

@claudeai Follow us on X- @Krama_AI

[Keyur Shah @keyur19198]

@claudeai Learn more- https://t.co/vvDCJdya6d
Follow us on LinkedIn- https://t.co/2mUo14wE7I

[Keyur Shah @keyur19198]

@claudeai Awesome to see @claudeai launching this capability. We at KramaAI, have an ambient version of this product which allows you to work without needing to start/stop recording! Hit me up if you want to learn more or try the product out.

[Keyur Shah @keyur19198]

There is no better time in history to be a builder than this week, it’s been insane fun 🤩

[Keyur Shah @keyur19198]

There is no better time in history to be a builder than this week, it’s been insane fun 🤩

[Keyur Shah @keyur19198]

Both Anthropic and OpenAI seems to be on a mission to get everyone addicted(in a good way) to continue using Fable 5 and GPT 5.6 at consumption levels unimaginable a year ago. My token burn in the last couple of days has easily been north of 100M tokens if not significantly more. I wonder what happens when breaks are applied and we no longer have these resets come in.

[Keyur Shah @keyur19198]

https://t.co/GmZUwioYwl

[Tibo @thsottiaux]

Another reset for our Codex and ChatGPT Work users. Actually hit 9M active users way earlier today, but then got distracted by the approximately millions of things the team is doing to keep the systems up and reliable.

Should have that sweet 100% weekly usage limit back in a few minutes. Go be your productive self and close twitter. Shoo!

[Keyur Shah @keyur19198]

@thsottiaux Got this message but the usage does not clearly communicate how much model capacity we are using similar to how Claude communicates Fable 5 usage. Bummer given I had no idea I was approaching capacity. It makes sense to have limits per model but visibility into usage and some form of warning will be great.

[Keyur Shah @keyur19198]

Implementation of GPT5.6 Sol is unbelievably good. /goal ran for 4 hours, tested, and executed the feature requirements with spot on pass quality towards product’s evals.

[Keyur Shah @keyur19198]

@thsottiaux Speed to feedback resolution is crazy, great to see such focus on improving product based on feedback from X 🤯
https://t.co/ROaiV2H3br

[Keyur Shah @keyur19198]

@thsottiaux What this means is essentially if someone is using 5.6Sol High to build a small feature, you essentially get 5 shots in a week as roughly one request consumes ~17% on an avg weekly usage. This does not happen when working with Fable5 High

[Aravind Srinivas @AravSrinivas]

there are two viable paths to overcome the power botteneck in data center inference:

1) local models orchestrating most of the token flow
2) solar powered data centers in space

[Keyur Shah @keyur19198]

@claudeai Let’s go 🚀🚀

[Keyur Shah @keyur19198]

Most enterprises need to go from 1. Having no visibility in how work happens in their organization to start capturing intelligence from workflows and 2. Do evals to transition towards an organization that is truly AI native. Most companies are in phase 1 still trying to figure out how to capture ground reality of how work happens which is where KramaAI is focused on.

[Satya Nadella @satyanadella]

https://t.co/xv6csf1SbV

[Keyur Shah @keyur19198]

Maxed out @claudeai 's Fable 5 usage twice on the Max plan this week. Started working on "fun" side projects to use up quota that would have just gotten wasted at 10PM yesterday night

What came out of it:

1. Family health records: photographed lab reports → local vision model (qwen3-vl on Ollama) → FHIR → self-hosted Medplum EHR. Oura + Apple Watch sync. Weekly trend digest for doctor visits.
2. Finance app: parses statements from 6 banks, tax lots + wash sales, Monte Carlo sims. Monarch/YNAB charge ~$100/yr and do less.
3. A learning coach that rewrites its own curriculum weekly.

All local-first. Health data never leaves my Mac — only de-identified numbers go to the Claude API.

~7,000 lines of Python in a day, most of it glue between great open source (Paperless-ngx, Medplum, Ollama). The real work was deciding what to build.

Strange, great time to know what you want.

[Keyur Shah @keyur19198]

@AravSrinivas @AravSrinivas do you also think that we have a high chance of seeing more vertical open source models or you think they will continue to remain closed?

[Keyur Shah @keyur19198]

@thsottiaux 🚀🚀🚀

[Keyur Shah @keyur19198]

Cranking in as much Fable 5 usage as possible before the July 12 cutoff. Have been able to get a lot done with Fable 5 vs unfortunately GPT 5.6 Sol ends up consuming too many tokens too quickly and need to wait for 5 hour limit to reset. Hope @ClaudeDevs shares one more news of reset this weekend :)

[Keyur Shah @keyur19198]

Great timing, need to work with 10% for rest of the week :/

[Claude @claudeai]

We're extending access to Claude Fable 5 on all paid plans through July 12.