I left Twitter for 10 years. Never becoming the story was the discipline. That rule breaks when machines and institutions will infer you from fragments anyway. I came back to leave a public record of judgment — not a growth account. Start here: dougstandley.com/essays/the-…

We’re in private beta. Get cosigned ✍️ to access and we’ll see you inside. 🤠

The more I knew, the less I understood. In the ’90s, I was the consultant who helped fire an entire leadership team in Mexico for fraud. Then I became El Jefe. The workers cried—not for the thieves, but for the men who gave them uniforms, meals, and buses. A year later I remembered Carlos, the hugs, and the whispers. Not a single KPI. In this essay, I try to describe what actually lasts. dougstandley.com/essays/the-… Hey, this is for you @Saul_Loveman and @mikeygnft!

For most of my working life, getting a good answer was expensive. You needed expertise, time, research, meetings, and often a room full of people. AI is changing that economics quickly. When a competent answer can be produced in minutes, the answer itself becomes less valuable as a signal. The interesting question becomes: Who knows which answer matters? And perhaps even more importantly: Who is willing to act on it? What do you think becomes scarce when answers become nearly free?

Silicon Valley folks who hate the labs and don't believe in x-risk keep uncovering the diabolical fact AI safety people use money and companies to get out their message, while the "just accelerate, it'll be fine crowd" involves such non-moneyed and anti-corporate interests as ... Nvidia, Andreessen Horowitz, and the pro-AI Leading the Future super PAC backed by tech executives.

The more I work with AI agents, the more I think the real challenge isn’t intelligence. It’s trust. Once an agent can read your email, access your files, manage your calendar, or take actions on your behalf, permissions become part of the product itself. I think the companies that get this right won’t just build smarter agents. They’ll build agents people are comfortable giving real responsibility to.

Running a multi-agent writing workflow with Claude Opus 5.5 + Codex Astra. I’m the human message bus. Found a shared bug: false authorship guardrails were suppressing useful contributions. Patch: protect my judgment, not every thought, from outside influence. “Hey, what if?” is back online.

I loved this part of the conversation with @tobi about finding interest in everything.

When we say there's an "alignment issue," how many of you mean this version?

We are becoming very good at asking whether an AI system can make a decision. I'm less interested in that question than I am in what happens afterward. Suppose the model makes the recommendation, the human approves it, the organization deploys it, and the outcome is harmful. Who actually owns the decision? The model cannot stand in the room. The organization may say the human approved it. The human may say the system recommended it. Everyone was involved. But involvement is not the same thing as responsibility. Where should responsibility live when intelligence is distributed across a system?

It's so incredible to see some people having 2-3 passion projects. Through those projects, they land tons of other things across the stack.

What’s surprised me most isn’t even that the forecasts were wrong. It’s how quickly the things I thought were years away keep moving into the let’s see what happens this year bucket. I’ve had to update my own expectations more times than I expected over the last year. And that’s probably the bigger lesson for me. When the underlying technology is improving this quickly, yesterday’s reasonable assumptions can become today’s biggest blind spot. I’m trying to spend less time asking could this really happen? and more time asking what would I do if it did?

This is pretty wild. Claude found a previously unknown enzyme system hiding in bacteriophage DNA, and it looks like it may belong to the same general family of programmable DNA systems that can cut, copy, and paste genetic material. We have no idea yet what this one actually does, which is probably the most interesting part. CRISPR started as a strange biological mechanism that researchers were trying to understand. It eventually became one of the most important tools in genetic medicine. Maybe this turns into something useful. Maybe it doesn’t. But having AI help researchers find biological systems we didn’t even know were there feels like a pretty important capability to have. I suspect we’re going to see a lot more discoveries where the first step is simply: Wait, what is this thing?

hello, world f13.com/

AI makes authorship strangely difficult to locate. If I give a model a prompt, it generates the language. If I edit the output, I influence it. If I approve the final version, I authorize it. But at what point did I actually author the thing? We have historically treated authorship as closely connected to production. AI may force us to separate production from authorship entirely. If the machine produces the words but the human chooses what deserves to exist, is that still authorship-or something else?

At Box, we've been testing Opus 5.5 on a variety of complex enterprise knowledge work tasks dealing with unstructured data with the Box Agent. Overall, we saw frontier capability levels, with major performance improvements over Opus 5. 63% fewer tokens used, 42% less verbosity, and 30% faster vs. Opus 5. And the model itself is cheaper, so this is a major win for any agentic computer use, coding, analytics, or data work that enterprises will be doing. Here are some examples of the task wins and performance gains across a variety of industry tests that we performed: • Financial services - due diligence (+39% task accuracy): A year of transaction records, with the job of finding every miscalculation in an acquisition target's pricing tool. Opus 5.5 scored a perfect result on every attempt in half the words Opus 5 used, consuming 82% fewer tokens overall. • Technology - cloud cost analysis (+65% task accuracy): Work out what a company should actually change about its cloud spend. Opus 5.5 picked the right basis for the retention calculation and kept the source data's unit conventions straight all the way through, so the number at the end actually holds up. It took half the time Opus 5 took, with 70% fewer tokens. • Consumer products - client account analysis (+17% task accuracy): Set the onboarding targets for a client account, reading across the signed contract, a satisfaction tracker and a team metrics sheet. The contract never states a senior/junior split, so Opus 5.5 derived it from the 18-person roster and showed the rule it used; several clients had a perfect 10 on individual survey questions, so it averaged each client's responses instead of crowning the single 10. It finished this one in half the time, on 78% fewer tokens. • Clinical diagnostics - data analysis (+15% task accuracy): Malaria rapid-test performance across a dry and a wet season: build the patient records out of two clinical PDFs, compute positive test rates by season and gender, and test whether parasite counts really differ between test-positive and test-negative patients. Opus 5.5 caught that the two groups' standard deviations differed more than 100-fold, re-ran it the right way, and found the dry-season difference didn't hold up after all. This accuracy gain came with a final answer that was half the length of Opus 5's, and also needed 78% fewer tokens end to end. Customers will be able to build AI Agents with Opus 5.5 shortly in the Box AI Studio.

Congratulations, @danieldines, @uipath, on The Work That Remains (free to download). "AI proposes. Humans decide. Automation executes." is the clearest operator's case I've read for the human at the gate. The question I keep coming back to: when that human says no, what makes the no stick? 👇👇👇👇

The 40% cost reduction is the part I’m really watching. Cheaper inference changes the product, not just the margins. You can run more agents, give them more context, let them take more shots at a problem, and automate things that didn’t make economic sense before. The best part of this race may be all the things we haven’t thought to build yet because the economics weren’t there.

Keep a human in the loop sounds like a safeguard. But a human being present is not the same thing as a human being in control. If the system chooses the options, frames the problem, determines what information is visible, and recommends the action, the human may only be approving a decision that has already been substantially shaped. So perhaps the real question isn't: Is there a human in the loop? It is: Where, exactly, does the human exercise judgment? That seems like a much harder question to answer.

This morning, me pressing a design decision: Do you really want @salesforce forever and @SlackHQ becoming a critical ungovernable utility forever?

The @X prompt window asks, What's Happening? What if we all answered right now? I'll go: My eyes are burning like hell from staring at this screen for eight hours, writing the narrative for a total corporate rebrand strategy. Who's next?