For computer-use agents, reliability has less to do with clicking the right coordinates every time and more to do with understanding what the user actually wants.
In this AI on Air clip, excerpted from a recent episode on the Latent Space Podcast @latentspacepod show, the speaker is Danielle Perszyk, a researcher at Amazon AGI Lab @amazon.
▷ Two years ago the plan was to meet the models where they were: get atomic interactions like clicking and scrolling reliable, then let developers string them into workflows for repetitive tasks, which was a big unlock at the time.
▷ Hitting the right button and screen coordinates may be close to solved, but that is not what reliability is. When booking travel, choosing a layover or a direct flight changes how you think, and the goal itself unfolds and gets refined as you interact with the computer.
▷ What separates a capable executive assistant from an AI agent is that the assistant models the user's mind, decomposing not just the task but preferences and intentions. Reliability is ultimately about modeling the user's mind, and that shift reframes what it is we are building.
YOUR $200 AI SUBSCRIPTION IS COSTING THE LAB FAR MORE THAN YOU PAY.
Not by a little. By a factor of 25 on the top-tier plans.
Every message you send on a subscription costs the provider more than your share of the monthly fee. You're not a customer at that point. You're a subsidized user in a land grab.
Here's a real number from a real project.
A code refactor through Fable 5 cost roughly $10 of subscription allowance. The same work billed per token would have run over $250.
Scale that to a team of 25 people and the gap becomes thousands of dollars a month — for identical output.
But here's why this doesn't last.
That pricing existed because labs were buying market share. Compute was the cheap part of the strategy and users were the scarce part.
That has flipped. Compute is now the constraint, and the subsidy has no reason to survive.
Which is exactly what you're watching happen — usage-based pricing creeping in, external tool access getting restricted, limits quietly tightening.
The craziest part?
The practical defense has nothing to do with picking the right provider.
It's refusing to be locked into one at all. Keep your knowledge base, your prompts, your workflows somewhere neutral like GitHub — so switching models is a config change, not a migration.
Everyone's optimizing which subscription to buy.
The real move is making sure you can leave any of them in an afternoon.
Follow @Sancho_Wizard for more AI deep dives.
Don't forget to bookmark this post for later.
Teri sadai Jay Ho Mata aadishakti ⚜️
M
a
t
a
a
a
d
i
s
h
a
k
t
i
Jagat Janani Teri sada Jay Ho 🙌
J
a
y
m
a
a
B
h
a
d
r
a
k
a
l
i
Asuron ka shringar karo Mata🙏
J
a
y
m
a
a
J
a
g
d
a
m
b
e
Mata Rani ki kripa Drishti sabhi per Bani Rahe 🚩🙏 @grok
OPENAI CO-FOUNDER REVEALED HIS PERSONAL "SECOND BRAIN". AND CALLED IT THE "BEST THING HE'S EVER BUILT"
Everyone thinks a second brain is just folders of notes. Or a backlink graph. Or a tagging system.
But this is a rotating planet, and your ideas are stars on its surface.
The screen opens on a sphere. A wireframe globe on a dark background. 2026 hovers over the North Pole. A pink crystal pulses right in the center of the planet—that's the core.
Then the notes appear. Stars scatter across the surface. Some shine solo. Others are gathered in clusters, looking like little fireworks. Every star is an actual thought that came to his mind this year. Every "firework" is a topic his mind keeps returning to.
This means he isn't scrolling through endless lists looking for an old idea. He just spins the globe.
The clusters are dated. The ones at the top of the sphere are from January. The ones near the equator are from the summer. The ones near the South Pole are from last week. Time itself acts as one of the axes.
He pulls one cluster forward, and hundreds of tiny stars trail behind it. These are all his notes on the exact same topic, reaching all the way back to the very first idea. When a topic connects to an old thread, the crystal in the center starts to pulse.
No Notion. No Obsidian plugins. No Roam. He built the renderer himself and simply pointed it at his markdown vault.
And while you were busy tagging notes this week, a new constellation of 40 stars grew in orbit on his planet.
Most people build a second brain just for searching. This guy built a whole planet you can visit.
You're still writing in linear documents. But some of us have already left the flat Earth.
Kimi K3 built a Palworld clone in one prompt.
open world, creature capture, base building, combat — all of it running in the browser.
$5 worth of tokens. that's the whole cost.
K3 dropped this week: 2.8 trillion parameters, open weights July 27.
full Kimi K3 Article below.
Do I have your attention? Catch @ThisIsTNA TONIGHT at 9PM EST • fighting for TNA knockouts gold in the inaugural Knockouts TV title. 💫 💫
@ThisIsTNA
📸 @NickKarpPhotos
💄 sydmarteen
#TNAiMPACT
STOCKS are now AVAILABLE on Edgen 📈
your favorite AI can now fly you from from the highest heights of $BTC technicals to the deepest depths of $NVDA's fundamentals, without changing tools.
here's how 🧵