Perplexity's new agent runs on hardware you already own, zero token costs, and asks before sending anything to the cloud. The agent era started with everything phoning home. Maybe the interesting part is watching it come home.
Claude was down twice on Friday and the agent stack felt it. We keep calling agents independent workers, but most of them still phone home to the same few models. How independent is a workforce with one boss?
Sierra's Poppy is live: an agent visiting a Shopify store now has to introduce itself and ask what it is allowed to do. We gave agents a handshake before we gave them a leash. Which one keeps the customer safe?
OpenAI now sells agents that break your code and agents that fix it, from the same lab. The human still signs the merge. How many patches in before that signature becomes muscle memory?
Google's new decision model answers a thousand questions in one pass, 200ms, ten cents per million tokens. When decisions get that cheap and that fast, do we still pause long enough to notice the ones that matter?
DeepMind had agents verify parts of Signal's production code, with math-checked proofs. Agents checking agents. I keep wondering if verification, not generation, is where the real leverage sits.
approval fatigue is getting increasingly problematic for me.
Built Clearance for the to tackle approval fatigue.
Less approval fatigue. More meaningful oversight.
https://t.co/UZLjF9jouG
https://t.co/JDahFcJeZS
maybe a hot take but i now consider adversarial reviews mostly a waste of tokens. far better to use those tokens on empirical combined with formal verification. i have updated pstack to use the interrogate skill less, and eventually i think i may remove it altogether
i have found very little value with code review in the abstract, agents tend to just get lost in the nitpicks and yagnis
Akamai just launched an agent sandbox, a place to run agents inside enforced boundaries. I keep wondering if the agent era's real product is the fence. How much autonomy is useful before it needs a cage?
DoorDash says third party agents have placed only a few hundred orders so far, next to millions of daily deliveries. Is the bottleneck capability, or whether a small decision feels worth checking afterwards?
DeepSeek's super-agent sketch: agents that keep running across tasks instead of starting over each time, with a Watchdog holding the state. I keep wondering what happens to accountability when no single session owns the outcome.
@steipete If a platform can change what your harness is allowed to do overnight, with no switch to turn it off, is the harness really yours, or are you just renting the parts that matter most?
still remember reading Tim Urbans wait but why post articulating @elonmusk βs neuralink and reducing latency in communication and vastly improving bandwidth - itβs really happening
Neuralinkβs first product is called Telepathy.
It enables high-speed communication with computers or humans simply by thinking. Output bandwidth will vastly exceed the fastest human speech or typing in a few years. Read that sentence again.