@smithstephen.bsky.social
New Setup Guides for Claude Team and ChatGPT Business. Anthropic changed its admin console so much that a good chunk of its own help articles are out of date. So I went through the live screens one setting at a time (and did the same things for ChatGPT!)
www.smithstephen.com/p/new-setup-...
www.smithstephen.com/p/new-setup-...
New Setup Guides for Claude Team and ChatGPT Business
Anthropic changed its admin console so much that a good chunk of its own help articles are out of date. So I went through the live screens one setting at a time.
www.smithstephen.com
September 28, 2026 at 8:37 PM
New Setup Guides for Claude Team and ChatGPT Business. Anthropic changed its admin console so much that a good chunk of its own help articles are out of date. So I went through the live screens one setting at a time (and did the same things for ChatGPT!)
www.smithstephen.com/p/new-setup-...
www.smithstephen.com/p/new-setup-...
To say that I have been blown away by the new Opus 5.5 model would be a serious understatement. It's better than any model I have used. Ever. Better than Fable. Better than Astra. So much for pacing the frontier! Here's a quick take as I've been on vacation for the past week.
Claude Opus 5.5 Is the Best Model (For Now!) I’ve Ever Used
I put three versions of the same deck in front of a friend on my trip. I didn't have to point out the gap.
www.smithstephen.com
September 27, 2026 at 12:05 PM
To say that I have been blown away by the new Opus 5.5 model would be a serious understatement. It's better than any model I have used. Ever. Better than Fable. Better than Astra. So much for pacing the frontier! Here's a quick take as I've been on vacation for the past week.
Anthropic is merging Claude chat and Cowork into one product so you no longer pick a mode. They also launched Claude Docs and Claude Slides, which go after Google Docs and PowerPoint directly. Here's my take on both, and the one setting a law firm has to get right before the merge reaches its people
Claude Stopped Asking Which Mode You Want. Your AI Policy Just Expired.
On September 16 Anthropic merged Claude chat and Cowork and launched Claude Docs and Slides in beta. Here's what changed, what it costs, and the one setting to rebuild your policy around.
www.smithstephen.com
September 17, 2026 at 4:12 PM
Anthropic is merging Claude chat and Cowork into one product so you no longer pick a mode. They also launched Claude Docs and Claude Slides, which go after Google Docs and PowerPoint directly. Here's my take on both, and the one setting a law firm has to get right before the merge reaches its people
Anthropic, OpenAI, Google and xAI all agreed within one day this week that AI should slow down, and the fix on the table is a federal testing law with the power to block models from the US market. Here's y view after 25 years in senior tech roles.
Four AI Rivals Agreed to Slow Down in One Day. Ask What Each One Gets
One of them is heading for the largest IPO in history. After July's attack, the company that got hit ran its forensics on a Chinese open model because Claude refused the job
www.smithstephen.com
September 14, 2026 at 3:05 AM
Anthropic, OpenAI, Google and xAI all agreed within one day this week that AI should slow down, and the fix on the table is a federal testing law with the power to block models from the US market. Here's y view after 25 years in senior tech roles.
Your Claude project takes "unlimited" files. It reads about 1,000 pages. Everything after that gets searched instead of read, and it won't tell you when it switched.
intelligencebyintent.com/insights/you...
intelligencebyintent.com/insights/you...
Your Claude Project Quietly Demoted Itself to a Search Tool | Intelligence by Intent
The answers keep coming back confident. They just stopped covering all 40 agreements, and nothing tells you which ones got left out.
intelligencebyintent.com
August 4, 2026 at 10:24 PM
Your Claude project takes "unlimited" files. It reads about 1,000 pages. Everything after that gets searched instead of read, and it won't tell you when it switched.
intelligencebyintent.com/insights/you...
intelligencebyintent.com/insights/you...
A managing partner's one worry about his new AI contract: locking in a price for something that gets cheap next year. Three weeks later, OpenAI cut prices 80%. My new article covers what his firm just gained without renegotiating a thing, and what yours did too.
The Price of Thinking Just Fell 80%. The Price of Judgment Didn't. | Intelligence by Intent
A managing partner asked me if he was locking in a price for something that gets cheap next year. It took three weeks.
intelligencebyintent.com
August 2, 2026 at 3:35 AM
A managing partner's one worry about his new AI contract: locking in a price for something that gets cheap next year. Three weeks later, OpenAI cut prices 80%. My new article covers what his firm just gained without renegotiating a thing, and what yours did too.
Nine hours a week. That's what your best people are saving with AI, and your firm is seeing almost none of it show up anywhere that matters. I mapped exactly where those gains disappear, and what the top 6 percent did to keep them: full piece below.
Your AI Saves Nine Hours a Week. Your Firm Sees None of It. | Intelligence by Intent
Four studies, 11,000 professionals, one finding: the individual time savings are real and the firm captures almost none of them.
intelligencebyintent.com
July 22, 2026 at 3:21 AM
Nine hours a week. That's what your best people are saving with AI, and your firm is seeing almost none of it show up anywhere that matters. I mapped exactly where those gains disappear, and what the top 6 percent did to keep them: full piece below.
The people going hardest at AI are the worst at being present. That's my crowd, that's me. There is no after, and the frontier doesn't pull over.
intelligencebyintent.com/insights/the...
intelligencebyintent.com/insights/the...
The Music Started, and Everything Else Went Quiet | Intelligence by Intent
My whole week is AI, a low hum of it running all the time. For two hours at the Bowl, I wanted nothing else.
intelligencebyintent.com
July 15, 2026 at 7:46 AM
The people going hardest at AI are the worst at being present. That's my crowd, that's me. There is no after, and the frontier doesn't pull over.
intelligencebyintent.com/insights/the...
intelligencebyintent.com/insights/the...
Everyone thinks AI is cutting jobs. Follow the money and the opposite shows up: the heaviest AI spenders got more productive, grew, and then hired. Here's the data, and what the winners did differently.
www.smithstephen.com/p/the-heavie...
www.smithstephen.com/p/the-heavie...
The Heaviest AI Spenders Are Hiring More People, Not Fewer
More than a third of American workers now use AI. Job openings held steady, and the heaviest spenders hired more, not fewer. So what's actually going on?
www.smithstephen.com
July 5, 2026 at 2:43 PM
Everyone thinks AI is cutting jobs. Follow the money and the opposite shows up: the heaviest AI spenders got more productive, grew, and then hired. Here's the data, and what the winners did differently.
www.smithstephen.com/p/the-heavie...
www.smithstephen.com/p/the-heavie...
Everyone's asking which AI is smartest. Wrong question. The one that matters: where should the work happen, and who checks it?
www.smithstephen.com/p/the-smarte...
www.smithstephen.com/p/the-smarte...
The Smartest Model Is the Wrong Question
If it ends up in a folder, hand it to an agent. If it's just a thought, stay in chat. And when your name's on the result, run it through both.
www.smithstephen.com
June 29, 2026 at 2:32 AM
Everyone's asking which AI is smartest. Wrong question. The one that matters: where should the work happen, and who checks it?
www.smithstephen.com/p/the-smarte...
www.smithstephen.com/p/the-smarte...
I asked a room of family lawyers to guess how much of the global AI hallucination problem was theirs. The lowest guess was 20 percent; the real number is 6. Most of those cases were filed by parents fighting custody alone with a free chatbot, and the new piece is about where that leads.
The AI Hallucination Problem Isn't Coming for Lawyers. It's Coming for the Courts.
I split 99 cases by who filed them. Most weren't lawyers. They were parents handling custody alone, trusting a free chatbot that handed them cases that never existed.
www.smithstephen.com
June 26, 2026 at 4:44 PM
I asked a room of family lawyers to guess how much of the global AI hallucination problem was theirs. The lowest guess was 20 percent; the real number is 6. Most of those cases were filed by parents fighting custody alone with a free chatbot, and the new piece is about where that leads.
Right now someone at your firm is pasting client facts into a free chatbot to save an hour. California's proposed AI rules might call that revealing confidential information. Do you know which tools your people are using?
www.smithstephen.com/p/california...
www.smithstephen.com/p/california...
California’s New AI Rules Don’t Create New Duties. They Just Took Away Your Best Defense.
The State Bar’s proposed changes make clear that competence, confidentiality, candor, and supervision already apply to AI. The “the rules don’t mention it” defense is gone.
www.smithstephen.com
June 24, 2026 at 4:56 AM
Right now someone at your firm is pasting client facts into a free chatbot to save an hour. California's proposed AI rules might call that revealing confidential information. Do you know which tools your people are using?
www.smithstephen.com/p/california...
www.smithstephen.com/p/california...
Your firm said no to ChatGPT and yes to Microsoft Copilot. Copilot runs the same OpenAI and Anthropic models, just a different contract. I read all three terms side by side, and the real risk isn't the vendor.
www.smithstephen.com/p/microsoft-...
www.smithstephen.com/p/microsoft-...
Microsoft Copilot Isn't Safer Than ChatGPT. It's the Same Engine in a Different Contract.
I read the commercial terms for all three, side by side. The vendor you pick barely matters. The account your people use is the whole game.
www.smithstephen.com
June 18, 2026 at 12:03 AM
Your firm said no to ChatGPT and yes to Microsoft Copilot. Copilot runs the same OpenAI and Anthropic models, just a different contract. I read all three terms side by side, and the real risk isn't the vendor.
www.smithstephen.com/p/microsoft-...
www.smithstephen.com/p/microsoft-...
A model showed up that stopped feeling like a smart associate and started feeling like a senior thinking partner. It lasted three days. The capability jump is real, the governance burden is real, and the firms that learn to hold both will be ready for what comes next.
The Best Legal AI Most Lawyers Will Never Get to Use Was Here for Three Days
The question is no longer whether models will get good enough for serious legal work. They already are. The only question is whether your firm can govern them before the next jump hits.
www.smithstephen.com
June 16, 2026 at 4:11 AM
A model showed up that stopped feeling like a smart associate and started feeling like a senior thinking partner. It lasted three days. The capability jump is real, the governance burden is real, and the firms that learn to hold both will be ready for what comes next.
You can't hand a model thousands of documents if it might tell you it read them all when it skimmed half. That's been the whole problem with AI on real matters. Opus 4.8 didn't fix it, but it moved in the right direction, and I tested it on a live 3,000-document matter the day it launched.
The Best Thing About the New Claude Isn't That It Got Smarter. It Got Honest.
You can't hand a model 4,000 documents and a deadline if it might tell you it read all of them when it skimmed half. That just changed.
www.smithstephen.com
May 29, 2026 at 12:30 AM
You can't hand a model thousands of documents if it might tell you it read them all when it skimmed half. That's been the whole problem with AI on real matters. Opus 4.8 didn't fix it, but it moved in the right direction, and I tested it on a live 3,000-document matter the day it launched.
The bar moved while everyone was arguing about whether to use AI. The attorneys closing the gap between using it and using it well are writing the rules the rest of the profession gets measured by
www.smithstephen.com/p/ai-doesnt-...
www.smithstephen.com/p/ai-doesnt-...
AI Doesn't Have a Bar Number. You Do.
A 100-document production used to eat a paralegal's afternoon. It took me ten minutes and a refilled coffee.
www.smithstephen.com
May 27, 2026 at 4:27 PM
The bar moved while everyone was arguing about whether to use AI. The attorneys closing the gap between using it and using it well are writing the rules the rest of the profession gets measured by
www.smithstephen.com/p/ai-doesnt-...
www.smithstephen.com/p/ai-doesnt-...
AI is horsepower. RevOps is the chassis, the brakes, the steering. Most companies are bolting 700 horsepower onto a chassis that was never built for it, and calling it strategy.
www.smithstephen.com/p/your-ai-st...
www.smithstephen.com/p/your-ai-st...
Your AI Strategy Is Just Your Dysfunction at 700 Horsepower
If your CRO and your CFO walk into the same board meeting with two different forecast numbers, an AI agent will not fix that. It will make it happen four times an hour.
www.smithstephen.com
May 26, 2026 at 2:14 PM
AI is horsepower. RevOps is the chassis, the brakes, the steering. Most companies are bolting 700 horsepower onto a chassis that was never built for it, and calling it strategy.
www.smithstephen.com/p/your-ai-st...
www.smithstephen.com/p/your-ai-st...
His IT director told him ChatGPT memory was off. He showed me the setting. He felt fine. Then the model summarized a contract by referencing another client.
www.smithstephen.com/p/he-turned-...
www.smithstephen.com/p/he-turned-...
He Turned Off ChatGPT's Memory. It Referenced Another Client Anyway.
One partner. Two clients. A setting that was supposedly off. Here's what's actually happening behind the toggle.
www.smithstephen.com
May 25, 2026 at 6:43 PM
His IT director told him ChatGPT memory was off. He showed me the setting. He felt fine. Then the model summarized a contract by referencing another client.
www.smithstephen.com/p/he-turned-...
www.smithstephen.com/p/he-turned-...
Andrej Karpathy. Anthropic. He could have gone anywhere. He picked the lab that's profitable, compute-rich, and winning the rooms that pay. That's the whole story this week.
www.smithstephen.com/p/karpathy-j...
www.smithstephen.com/p/karpathy-j...
Karpathy Just Picked Anthropic. That's the Whole Story.
He could have gone anywhere in AI. He didn't. Here's what that one decision tells you about the bets your firm is making right now.
www.smithstephen.com
May 22, 2026 at 8:17 PM
Andrej Karpathy. Anthropic. He could have gone anywhere. He picked the lab that's profitable, compute-rich, and winning the rooms that pay. That's the whole story this week.
www.smithstephen.com/p/karpathy-j...
www.smithstephen.com/p/karpathy-j...
Forget faster models. Google shipped a worker. Spark runs on its own VM, 24/7, with your laptop closed. The chatbot phase is over. I/O 2026 made it official.
www.smithstephen.com/p/google-jus...
www.smithstephen.com/p/google-jus...
Google Just Skipped the Chatbot Era. Your 2026 AI Plan Is Already Outdated.
The model isn’t the product anymore. The agent is. And Google now owns more of the surfaces it runs on than anyone else.
www.smithstephen.com
May 20, 2026 at 3:32 AM
Forget faster models. Google shipped a worker. Spark runs on its own VM, 24/7, with your laptop closed. The chatbot phase is over. I/O 2026 made it official.
www.smithstephen.com/p/google-jus...
www.smithstephen.com/p/google-jus...
I watched a video this weekend on YouTube and I cannot stop thinking about it. "Humans were the horses." Ken Griffin opened his Stanford answer on AI with that line, quoting Niall Ferguson. The 19 minutes that followed got more uncomfortable, not less.
Humans Were the Horses. Ken Griffin Isn't Hedging.
The work you thought was safe isn't. The kids you thought would be fine aren't. And the moats you thought protected you just got filled in.
www.smithstephen.com
May 17, 2026 at 4:46 AM
I watched a video this weekend on YouTube and I cannot stop thinking about it. "Humans were the horses." Ken Griffin opened his Stanford answer on AI with that line, quoting Niall Ferguson. The 19 minutes that followed got more uncomfortable, not less.
A general counsel emailed me last week: "Why are we suddenly getting bad answers from Claude?" They weren't. They were getting entry-level answers from a model they didn't pick, because the app had quietly reset their default. The labs need to fix this.
You Picked the Premium AI. The App Switched It Back.
It happened to four of my clients in two weeks. The picker quietly flipped, and nobody noticed.
www.smithstephen.com
May 7, 2026 at 5:21 AM
A general counsel emailed me last week: "Why are we suddenly getting bad answers from Claude?" They weren't. They were getting entry-level answers from a model they didn't pick, because the app had quietly reset their default. The labs need to fix this.
Most firms plan AI training as if every role moves at the same speed. The reality: 18 months for assistants, 24 for paralegals, 36 for lawyers. Get that spread wrong and you mis-staff every part of the firm at once.
www.smithstephen.com/p/your-assis...
www.smithstephen.com/p/your-assis...
Your Assistant Has 18 Months. Your Paralegals Have 24.
Most firms are planning training, hiring, and pricing as if every role is on the same curve. They're not. Here's what that mistake costs.
www.smithstephen.com
May 5, 2026 at 4:57 AM
Most firms plan AI training as if every role moves at the same speed. The reality: 18 months for assistants, 24 for paralegals, 36 for lawyers. Get that spread wrong and you mis-staff every part of the firm at once.
www.smithstephen.com/p/your-assis...
www.smithstephen.com/p/your-assis...
I asked Claude to build one PowerPoint template from a design spec it created. It failed three times. Codex did it in one shot, and that single test reshaped how I think about every AI tool decision I make.
www.smithstephen.com/p/i-tried-to...
www.smithstephen.com/p/i-tried-to...
I Tried to Build One PowerPoint Template in Claude. It Failed Three Times.
Three failed attempts in Claude. One pass in Codex. Here's what that taught me about every AI decision I've made since.
www.smithstephen.com
April 28, 2026 at 1:11 AM
I asked Claude to build one PowerPoint template from a design spec it created. It failed three times. Codex did it in one shot, and that single test reshaped how I think about every AI tool decision I make.
www.smithstephen.com/p/i-tried-to...
www.smithstephen.com/p/i-tried-to...
Norway's $2.2 trillion fund put their AI playbook on camera. What they showed first wasn't the technology.
A $2.2 Trillion Fund Showed Exactly How They Do AI. Steal This.
Norway's sovereign wealth fund published their full AI playbook. The lessons start well before the technology.
www.smithstephen.com
April 2, 2026 at 7:14 AM
Norway's $2.2 trillion fund put their AI playbook on camera. What they showed first wasn't the technology.