The famous Greek philosopher Heraclitus once said "there is nothing permanent except change." I'm 97.4% sure he was talking about this exact moment in time. Specifically, I can only assume he had the forethought to know that Anthropic would publish this article about AI building itself.

As the pace of change grows faster, so does the need for flexibility. A year ago, touching your toes was fine. But a year from now, the people who can't get a palm on the ground may get left behind.

That's the whole point of this issue. There's a single chart in that article that tells you everything about where we actually are, and it changes what "adopting AI" should even mean for your business.

This week: the deep dive on why "the pace is slowing" is the most expensive wrong belief in business right now, Claude as the lawyer who reads your contract before you sign it, and three stories worth your time (voice AI grows up, image editing gets surgical, and your coding bill goes on a meter).

On Thursday, Anthropic put out a piece called "When AI builds itself." The headline ask is dramatic: the company wants the world to build a coordinated "pause button" so that governments and AI labs could agree to slow down frontier development if things get scary, the same way nuclear powers built arms-control regimes. Take a breath, let society catch up.

That's the story most outlets ran with. Bloomberg, the works. It's worth reading, and I could talk about it and its implications for hours.

But the pause button isn't what I want to talk about here. The thing I keep coming back to is one chart. It's a bar chart of how much code each Anthropic engineer ships per quarter. For four years, 2021 through 2024, the bars are basically flat. One human, one normal amount of work. Then in 2025 the line starts to climb. Then in 2026 it goes nearly vertical. The most recent bar sits at 8x the old average. Eight times the output per person.

Here's the chart:

Now, here's the trap. A lot of smart people look at the last couple of years and say "the pace is slowing." ChatGPT in 2022 felt like a lightning strike. Most releases since have felt a little less jarring. So the brain files it under "the exciting part is over."

I think that's exactly backwards. The jumps feel smaller because we got used to them, not because they got smaller. A chart that bends straight up is not a chart that's calming down. The shock wore off. The curve didn't.

So what do you actually do with that?

This is where I want to be honest instead of sell you a tidy answer. The truth is the only real certainty right now is deep uncertainty. You can find a serious, credible person predicting almost anything. One camp thinks self-improving AI means we'll need universal basic income within five years because there won't be enough jobs to go around. Another camp thinks we hit a wall, a data wall or an intelligence ceiling, and we just keep getting slightly better chatbots without the sci-fi outcome. I have my own view (I don't think we hit the wall). But here's the uncomfortable part: the steps you should take for your business look completely different depending on which one is right.

So the worst move isn't betting on the wrong future. The worst move is betting your whole company/life on any single future. The business owner who fires the team and goes all-in on AI is making a bet. The business owner who ignores AI and says "we're fine" is making the opposite bet. Both are betting the house in a game where nobody can see the cards.

The actual play is learning how to place your palm on the ground… or rather, flexibility. There are clear, boring, real efficiencies sitting in your business right now, and you should go grab them. That part isn't uncertain at all. But the businesses that "adopt AI" once, check the box, and say "okay, we're good" are the ones that struggle. Not because they did the wrong thing. Because they did it once and stopped.

"Be flexible" is easy to say and useless on its own, so here's what it actually looks like in practice. Three things, none of which require you to predict the future.

First, rent, don't marry. When you bring AI into a workflow, favor month-to-month tools and reversible changes over multi-year contracts and deep custom builds welded to a single vendor. The last 10% of optimization you'd get from locking in is not worth losing the ability to switch when something better shows up in four months, and at this pace, something better will. Optionality is the asset right now.

Second, put one person on "what changed this month." Give someone a couple of hours a month to track what's new and, just as importantly, to re-test the tools you ruled out earlier. The thing that couldn't handle your task in January can often handle it by June. If nobody's job is to notice that, you'll keep running last quarter's playbook on this quarter's technology. Flexibility isn't a mindset, it's a recurring slot on someone's calendar.

Third, don't restructure your headcount around a forecast. Resist the urge to fire or hire purely on a bet about where AI is going. Keep humans in the loop and let the work grow into the tools, instead of betting your org chart on a future nobody can actually see yet. The reorg is the one move that's genuinely hard to undo, which is exactly why it shouldn't be the one you make on a guess.

Adopting AI in 2026 isn't a project with an end date. It's building the muscle to change fast, because the ground is moving faster than it ever has and the people who survive are the ones who can pivot when it does.

That's the whole reason Bridge AI exists. We don't drop in a tool and disappear. We embed, we build the muscle with your team, and we stay flexible alongside you. If "we adopted AI last year, we're set" sounds like your company, that's worth a conversation.

Every business owner has signed something they didn't fully read. An NDA before a partner call. A vendor agreement the other side sent over "just standard stuff." A 40-page contract the business team wants signed by Friday. You skim it, you feel a little nervous, and you sign anyway because hiring a lawyer to review a routine doc costs more than the doc is worth.

Anthropic recently released a free "Legal" plugin for Claude Cowork, and it's built for exactly this. You drop a contract into a folder, run one command, and Claude walks through it clause by clause. It flags everything with a simple traffic-light system. Green is fine, yellow is "look at this," red is "do not sign this as written." It hands you specific redline suggestions and plain-English explanations of why a clause is a problem.

For one of our clients, this turned a contract-review bottleneck that used to eat days into something closer to an afternoon. NDAs that used to sit in a pile get sorted in minutes, and only the real problem ones go up to a human.

To be clear, this does not replace your lawyer, and for anything high-stakes you still want one. Claude is the gut-check that catches the obvious landmines before the expensive review, not instead of it. A human still makes the final call. But "an experienced second set of eyes on every contract before it hits my desk" used to be a luxury. Now it's a download. If you're in Cowork, you can grab the Legal plugin and start using it today.

Voice AI quietly grew up, and almost nobody noticed

While everyone argues about which chatbot is smartest, the speech side of AI has quietly been getting very good very fast. At its Build conference last week, Microsoft shipped a wave of in-house "MAI" models, including a new transcription model that now handles 43 languages and can turn an hour of audio into text in under 15 seconds. The part that matters for businesses isn't the benchmark, it's where it landed: straight inside Copilot, Teams, GitHub, and its contact-center product. It's the same story across the industry, with OpenAI also pushing real-time voice models built to actually hold a live phone conversation.

My take: of all the AI categories, speech is one of the easiest for a normal business to actually plug in and put to work. The phone gets answered, a human-sounding agent responds, the call gets transcribed, the follow-up happens on its own, and increasingly it's baked into tools you already pay for instead of forcing a new workflow on your team. I think this is the area that quietly explodes for small and mid-size businesses next, not because it's flashy, but because it's so easy to implement.

READ THE BUILD RECAP →

Image editing stopped being a slot machine

Reve just released a new version of its image model that edits images the way you'd edit a document, not the way you'd pull a slot-machine lever. Instead of re-rolling the whole prompt and praying, it treats the picture almost like code, so you can change one specific part, the lighting, an object, the text on a sign, while everything else stays put. Shadows update correctly, perspective holds.

My take: the interesting thing here isn't Reve specifically. It's that a smaller company, not one of the giants, is the one pushing the boundary on precise editing. My guess is the big players copy this within months. But it's a good reminder that the frontier isn't only being moved by the names you know, and for any business doing its own product photos or marketing images, "change just this one thing" is a much bigger deal than "make me a new image."

READ THE REVE 2.0 BREAKDOWN →

Your AI coding bill just went on a meter

On June 1, GitHub switched Copilot, one of the most popular AI coding tools in the world, from a flat monthly subscription to usage-based billing. Instead of one predictable fee, you now burn "AI credits" based on how much you actually use. Some developers reported torching days' worth of credits in a couple of hours.

My take: this isn't really a coding story, it's a preview of where a lot of AI pricing is probably headed. The "all you can eat for $20 a month" deal was always a bit of a fiction, because running these models genuinely costs money. So the practical lesson is the opposite of "wait and see." Take advantage of the flat-rate pricing while it lasts. Hopefully it sticks around for a while, but nobody knows how long, so act now: get your team deep into the AI subscriptions you already pay for and learn to squeeze every dollar out of them while the meter is still off. The companies that build the habit now are the ones who won't flinch when the pricing eventually changes.

READ THE GITHUB BLOG REPORT →

That's it for this week.

If you found this useful, forward it to someone who keeps hearing "AI is slowing down" and believes it. They're the person this issue was written for.

And if you want to talk about what AI could actually do for your business, not in theory, but in your workflows, with your team, on your timeline, just reach out at [email protected]. That's what we're here for.