alfalama @blekachu - Twitter Profile

working on a deep research skill for codex remind me to share it tomorrow, i will leave it in a self improving /goal loop for the rest of the night hopefully i wake up to a banger might be trash we will find out in the morning gn 😴

5

70

4

14

3K

0

676

alfalama

@blekachu

13 days ago

Codex /goal idea you can try, I'm running it now: request all your data from X, then give the export to codex and ask it to do this, customized for you, for example around design, etc.: Go through my liked items from the last 3 months. Follow this process: 1 - Filter only items related to programming, products, AI, startups, design, business, and tools. Ignore memes and pure entertainment. 2 - Download the relevant items locally, including URL, author, date, and context. Go through the comments too. 3 - For articles, read the full text, not just the summary. 4 - For each strong item, research additional context: repo, docs, author, company/project, similar tools, and practical use cases. 5 - Finally, give me a ranked shortlist of the highest-value items: title, link, why it matters, what I should take away from it, and any suggested next step.

blekachu's tweet photo. Codex /goal idea you can try, I'm running it now: request all your data from X, then give the export to codex and ask it to do this, customized for you, for example around design, etc.:

Go through my liked items from the last 3 months.

Follow this process:

1 - Filter only items related to programming, products, AI, startups, design, business, and tools. Ignore memes and pure entertainment.
2 - Download the relevant items locally, including URL, author, date, and context. Go through the comments too.
3 - For articles, read the full text, not just the summary.
4 - For each strong item, research additional context: repo, docs, author, company/project, similar tools, and practical use cases.
5 - Finally, give me a ranked shortlist of the highest-value items: title, link, why it matters, what I should take away from it, and any suggested next step.

0

1

0

54

alfalama

@blekachu

13 days ago

It's honestly crazy what's possible with the latest long-running goals and capabilities of the models, where you can just go scrape a competitor's website, all the help docs, roadmap, or feature requests, transcribe them from YouTube videos, and basically map out a whole product. Combine it with karpathy-style wiki and boom.

0

46

alfalama

@blekachu

13 days ago

Not sure if I should tokenmaxxxx or tokenoptimizzzze

0

33

alfalama

@blekachu

13 days ago

First wedding planned in Kruspin

0

3

0

75

alfalama

@blekachu

14 days ago

If Codex keeps telling you that browser or computer use is not available in the thread during tool call, just tell it in a new thread to look through the historic threads and fix this issue so that it's always available in every thread.

blekachu's tweet photo. If Codex keeps telling you that browser or computer use is not available in the thread during tool call, just tell it in a new thread to look through the historic threads and fix this issue so that it's always available in every thread. https://t.co/GKE7KDsbhk

0

3

0

173

alfalama

@blekachu

14 days ago

First wedding tomorrow managed in my app. Let's goo. I'm on site checking for bugs, polishing and features.

0

3

0

63

alfalama

@blekachu

15 days ago

@kr0der https://t.co/pHU75Ljrei

Marty Kausas

@marty_kausas

16 days ago

Nobody has figured out how to price AI. Cursor just moved Bugbot from $40/seat/month to usage-based pricing at ~$1-1.50 per run. They realized seat-based doesn't work when an agent does variable amounts of work per user. Salesforce has changed Agentforce pricing 3 times in 18 months. → Started at $2 per conversation → Added Flex Credits at $0.10 per action → Added per-user licenses at $125/user/month → Now they offer ALL of the above and let customers self-select In the first two quarters, they closed 5,000 Agentforce deals but only 3,000 were paid. Customers wanted to try it but couldn't commit to a pricing model they didn't understand. Meanwhile, newer AI companies like Paper are still launching with plain per-seat pricing ($20/user/month). Even for a product built around AI agents. Everyone is experimenting. No one has converged. The three models I'm seeing... 𝗣𝗲𝗿-𝘀𝗲𝗮𝘁: Simple to sell, but disconnected from value. If one user runs 100 agent tasks and another runs 2, same price. Doesn't make sense long-term. 𝗨𝘀𝗮𝗴𝗲/𝗰𝗿𝗲𝗱𝗶𝘁-𝗯𝗮𝘀𝗲𝗱: Aligns with value, but creates anxiety. Customers don't know what they'll spend. And it can blow up on both sides. We experienced this firsthand. We bought a GTM product with credit-based pricing. We burned our entire expected yearly allotment in two months. We were unhappy. The vendor had to replenish our credits at no cost to keep us. Nobody won. 𝗢𝘂𝘁𝗰𝗼𝗺𝗲-𝗯𝗮𝘀𝗲𝗱: The dream. "Pay when the agent actually does the thing." But incredibly hard to define what a "successful outcome" even is. Who decides? The vendor? The customer? I think we're going to end up somewhere in the middle. Some base fee, plus usage on top. Basically how cloud infrastructure already works. But honestly no one knows yet. We're figuring this out at Pylon too. It's one of the harder problems we're working through as we roll out more AI products.

2

12

0

2

640

0

1

0

31