Software Development
levelsio highlighted xAI’s plan to supply tens of thousands of GPUs to Cursor for training its Composer 2.5 AI coding model, noting it as a strategic shift for cloud services to developers.
Oh yes oh yes 😍 https://t.co/AI5e2Fxpid
— @levelsio (@levelsio) April 16, 2026
🚨 VOID is now live on fal!
🗑️ Video object removal that also corrects physical interactions
💥 Remove a person or an object
🎬 Handles multiple objects, fast motion and complex backgrounds pic.twitter.com/xjAr5e1IUJ— fal (@fal) April 16, 2026
He also shared updates on Day 14 of the Cursor-sponsored #vibejam, praising AI-assisted game development entries like Capybara Delivery Driver and Floppy Brawler, with $40k in prizes and 15 days left.
🧑🚀 Day 14 of the @cursor_ai #vibejam
Proudly sponsored by @cursor_ai + @boltdotnew + @heyglif + @tripoai
Prizes to win (submit your vibe coded game before May 1!)
🏆 $25,000
🥈 $10,000
🥉 $5,000Damn I'm not saying to pump but today I felt the games are finally getting good… https://t.co/RgqPVUNNQn pic.twitter.com/hrmXcU1Nmo
— @levelsio (@levelsio) April 16, 2026
Automation & Orchestration
Discussions highlighted ongoing challenges in LLM context management, with suggestions that markdown files incorporating structured data and graphs could address this bottleneck for agentic systems.
context management is still LLMs' biggest bottleneck today…
are markdown files with structured data and graphs the solution?
— Machina (@EXM7777) April 15, 2026
It is notable that we are all debating exactly which markdown files are most important to feed AI (skills, memory, tool instructions) and in which order to feed them to get the best output. Feels that this is likely a temporary state of affairs in the development of agents
— Ethan Mollick (@emollick) April 12, 2026
There was also commentary on AI revenue agents for sales, critiquing much of the category as prompt engineering integrated with tools like Salesforce OAuth.
me realizing the entire "AI for sales" category is just prompt engineering with Salesforce OAuth https://t.co/DSc9FeaWEb pic.twitter.com/jKk0wviRJr
— Machina (@EXM7777) April 15, 2026
fal congratulated Nous Research on launching Tool Gateway in Nous Portal, integrating tools like web scraping, browser automation, image generation, TTS, and cloud terminals under one subscription to power agentic applications.
Nous Research🤝fal
Congratulations to @NousResearch on the Tool Gateway launch. Developer-first infrastructure is what will define the next phase of agentic applications. https://t.co/nahDROBbka
— fal (@fal) April 16, 2026
Strategy & Ecosystem
Key trends included the failure of predictions about an AI compute bubble or glut, with surging demand debunking recession fears from unused capacity.
Six months ago, there was a lot of focus on the idea that the there would be a massive glut of unused computing power which would could a recession as AI use plateaued. The "compute bubble" belief was absolutely everywhere.
The degree to which this was wrong deserves some notice
— Ethan Mollick (@emollick) April 13, 2026
Adoption progress was noted in ChatGPT’s user base, where the initial 80% male gender gap has fully closed.
Wish there was information about where this data came from, but this is a very significant change. Since AI use comes from experience, the persistent gender gap in AI use across every study of AI was something that a lot of scholars were concerned about. https://t.co/HkOTo71mT0
— Ethan Mollick (@emollick) April 15, 2026
Economic metrics for AI emerged, such as a proposed FLOP standard for inference costs, estimating $1 buys about 10^17 managed-LLM FLOPs.
Instead of the gold standard, we can imagine an inference standard of exchange, the FLOP. (As opposed to tokens, this accounts for AI ability)
With some AI help, I figure $1 buys roughly 10^17 managed-LLM inference FLOPs.
So that $4 coffee would cost half an exaFLOP, choom.
— Ethan Mollick (@emollick) April 15, 2026
Debates touched on automation’s job impacts, comparing opposition to Waymo’s self-driving tech to resisting medical advances.
This is like being against curing cancer because it will put oncologists out of work https://t.co/NJRXoChMvi
— Justine Moore (@venturetwins) April 13, 2026
Suggestions included more human-like model names like “Claude” for easier reference and anthropomorphization.
OpenAI should probably bite the bullet and just name their next set of models something more human sounding.
Everyone anthropomorphizes their AIs anyway, and "Claude" is an easier name to refer to than ChatGPT. Also easier to make a gerund, "Clauding," or adjective, "Claudy-y."
— Ethan Mollick (@emollick) April 12, 2026
Dan Shipper discussed a “philosopher draft” where AI labs hypothetically hire historical thinkers, sharing results in an image.
we ran a philosopher draft
which philosopher from history would each model lab hire if they could, and why? pic.twitter.com/KDryCy15V1
— Dan Shipper 📧 (@danshipper) April 16, 2026
He reacted positively to White House plans granting US agencies access to Anthropic’s Mythos, speculating it could secure contracts under Trump.
it would be incredible if the mythos announcement was a 4d chess move to make Trump keep the anthropic contracts
bullish https://t.co/yD5i2ztgrU
— Dan Shipper 📧 (@danshipper) April 16, 2026
Guillermo Rauch (@rauchg), CEO of Vercel, promoted their AI Gateway’s fixed pricing model as a response to criticism of Openrouter’s 5.5% fee on model requests, linking to the pricing documentation.
— Guillermo Rauch (@rauchg) April 10, 2026
Steal this idea.
Openrouter but with a fixed fee instead of extra 5.5% they charge.
Having a single interface for all models is amazing, but charging variable cost for proxying requests seems excessive.
Why not charge on per-user basis instead of 5.5% of all spend?
— Paras Chopra (@paraschopra) April 10, 2026
Creative & Visual Media
Seedance 2.0 generative video capabilities were extensively showcased, recreating historical scenes like Raphael’s School of Athens, Goya’s Saturn with hot dogs, and a mech battle between Neanderthals and Homo Sapiens.
All is not lost. Duckerton is still possible.
Here is Seedance 2.0 with the same prompt. https://t.co/Drunmv7ZNa pic.twitter.com/ZBLO70QGCE
— Ethan Mollick (@emollick) April 10, 2026
AI finally lets us see Raphael's The School of Athens the way Raphael obviously intended it, illustrating the delicate dance and subtle conflicts between Plato and Artistotle.
(Seedance 2.0 is very fun to play with) pic.twitter.com/YD7vVaRkFt
— Ethan Mollick (@emollick) April 10, 2026
I also fixed Goya's Saturn Devouring His Son. I assume the Black Paintings only existed because Goya did not have access to delicious hot dogs. pic.twitter.com/dwNKN3lHLz
— Ethan Mollick (@emollick) April 11, 2026
Impressed that Seedance 2.0 can pull of "a mech battle between Neanderthal and Homo Sapiens" so well. (This is exactly what happened, historically) pic.twitter.com/isz8yANi3k
— Ethan Mollick (@emollick) April 13, 2026
Emphasis was placed on scalable content creation, exemplified by AI-generated “fruits slop” videos earning $10k/month on TikTok by following trends and building efficient systems.
while you're working on your new revolutionary idea…
someone is making $10k/month running AI fruits slop on tiktok
they followed a simple trend, built an efficient system, and scaled it
you're still trying to become the next Elon Musk
maybe there's a lesson here about… https://t.co/RxRd7DdjsP
— Machina (@EXM7777) April 15, 2026
fal announced VOID video inpainting live on their platform, enabling object removal with physical interaction correction for multiple objects, fast motion, and complex backgrounds.
🚨 VOID is now live on fal!
🗑️ Video object removal that also corrects physical interactions
💥 Remove a person or an object
🎬 Handles multiple objects, fast motion and complex backgrounds pic.twitter.com/xjAr5e1IUJ— fal (@fal) April 16, 2026
They also launched ERNIE Image LoRA training for custom styles, characters, and palettes to maintain brand consistency.
🚨 ERNIE Image LoRA training is now live on fal!
🎨 Lock in your brand look, characters and recurring props
✨ Train custom styles and palettes your base model can’t ship out of the box
⚡ Fewer retries: personalization that stays on-model across campaigns pic.twitter.com/Dr8dhMR1zp— fal (@fal) April 16, 2026
AIWarper called for low-VRAM optimizations of the open-sourced Fish Audio S2 Pro TTS model and praised video-to-video (v2v) for AI filmmaking.
How has Fish Audio S2 Pro been fully open sourced for a month now and none of you have gotten this thing to work on low VRAM yet?
Seriously incredibly TTS model that no one is doing anything with 😂
Link below – plz make werk
— A.I.Warper (@AIWarper) April 16, 2026
Dope
I’ve always said v2v was the path forward https://t.co/5ngSuRmepe
— A.I.Warper (@AIWarper) April 16, 2026