Phil

Claude 5.5 feels like the last major llm version where development experience still gives you some leverage over non technical people. Claude Fable 6 will delegate most developers to product managers

Had to also make one opus5.5 video

I love 5.6-sol, it works for hours oneshotting tasks that completly stumped opus. But the pricing math just does not add up for 95% of enterprise customers. Most limit their employees to $1-2k per month but this 5.6 now burns $1k/day. It does not even matter how good the model is, everyone is still working 8h/day, irrespective to how productive. Also one enterprise user is now paying 100x more per month compared to one private pro $200 user. For the same product, just with no zero data retention option. Not quite sure what, but something will have to give here

Holy they still got it

Github Copilot CLI is the most underrated agentic coding tool out there, especially if you like using subagents. Being able to launch a fleet of subagents with whatever frontier models you like + this not costing extra credits has completely changed my workflow

I have tried a bunch of these projects, but after having codex run on my Desloppify for 8h straight I have to say, this is by far the best. The code has gotten so much better, stuff I would have never touched is cleanly refactored, everything works. Huge props

>new model out >dam it just one shots everything, never shipped so fast >it’s adding esoteric performance tweaks I don't even understand. we are so back >codebase growing 5k locs per day >what? model makes more mistakes now?! >they clearly quantized it in the backend >:(

AI Slide Generator Tier List: Great: Claude[.]ai - most flexible, consistent formatting, pptx export. Requires some iterations (slight overlap in texts) but only one I would use for work. Great at adding data visualizations and diagrams as well. Allows upload of templates for company specific requirements. Requires good prompting to not get color slop (img 1 as example). Powerpoint export possible. Good: NotebookLM - makes absolutely beautiful slides (using nano banana pro), super consistent and great styling. Deep research integration is nice. Downsides: not editable, only PDF export, every page has a NotebookLM logo. Best zero shot slides for usecases that don't need branding. (img 2 as example) Usable: Gemini App - decent formatting, export to Google Sheets, only one that actually uses images which fit (eg architecture images from the arxiv paper directly). No diagram support, not very flexible (img 3) Kimi - default templates, decent layout structure, no diagram support and random images. But powerpoint export so good enough for a starting point to fix it yourself. (img 4) Meh: Z[.]ai/GLM 4.6 - similar to Claude, but worse. Uses random images, slides are not consistent in length and width (which is a dealbreaker for me), but decent styling. Not exportable at all, basically just a website built in html. Gamma - uses pre-built components, looks okay at best, slides not consistent in size, AI generated images everywhere, annoying pricing structure. Seems to be targeted at middle school students, probably fine for that. Manus - model just not good enough to create high quality slides meaning weak content, slide sizes slightly inconsistent, powerpoint export at least but too many issues for manual fixing imo. (caveat I used Manus 1.5 Lite since I don't have the subscription) Unusable: minimax agent - did not even create slides for me, worked forever and created a markdown file in the end Runnable - random page sizes, terrible AI image generated diagrams with unreadable text everywhere, unredeemable currently (uses Claude but the scaffold is bad)

The day Anthropic teaches Opus to read my existing code before guessing what my DTOs look like

The main difference between codex-max and opus is that codex expects a fleshed out feature description to get good results while opus figures out what you want based on a vague non technical prompt. One isn't better, just personal preference on how you work and how much you want to be in control

Learned something new today. Opus is good

nano banana pro can visualize company earnings with its integratet search

nano banana pro about ai twitter

Sadly Antigravity stucks rn. I like the idea of the artifacts in the chat & the ability to leave comments on the implementation plan, but for some reason the model just edits random stuff, deletes code that should not be touched etc. Cursor with Gemini 3 works way better for now

hmm this looks familiar. now we know why google hired the senior windsurf staff

Gemini 3.0 will roll out on Google Antigravity? Seems like a new product is also launching today.

gemini 3 is about to cause absolute carnage in the ai coding tools market if google keeps those 1000 free requests imagine charging $200 when google drops the actual best model for free

This is to scale btw, and only includes burgers sold by McDonald's versus all of OpenAI's water consumption. If ppl were seriously concerned about fresh water, for every "OpenAI uses too much water," there should be 513 tweets complaining about beef production

Anthropic seems to be thinking about changing the Max 20x pricing structure, also moving to a credit based system