This Week in AI, in Five Minutes
This Week in AI, in Five Minutes
Every small business I know has the same junk drawer problem. Except instead of rubber bands and dead batteries, it’s software. One app runs the front desk, another handles reviews, a third schedules social posts, and after a couple years nobody on the team remembers which tab does what anymore. This week, the biggest AI company on the planet did something about that exact mess inside its own product line. That’s where we start.
CLAUDE COMBINED EVERYTHING INTO ONE APP
What’s Happening: Anthropic merged its two separate products, the regular Claude chat and the newer Cowork tool, into a single interface this week. Pro and Max subscribers got early access, with free and team plans rolling out over the coming weeks. The same window now handles quick questions, background research, documents, and slide decks, routing each request to the right mode without you picking a tab.
Why This Matters: If you’ve put off using AI because keeping track of separate apps felt like a hassle, that excuse is gone. Open one window, ask for what you need, whether that’s a patient newsletter draft or Monday’s team slide deck, and it figures out how to get there. Bigger picture, this is where every major AI company is headed: fewer tools, one assistant that actually adapts to what you’re trying to do. That’s good news whether you run a two-chair practice or you’re just trying to keep your own family’s calendar straight.
CLAUDE JUST BROKE A MATH RECORD THAT STOOD FOR YEARS
What’s Happening: Claude spent part of this month working on pure number theory and found new elliptic curves with a mathematical rank of 30 and 31. Human mathematicians had spent years pushing the previous record from 28 to 29, and Claude found two higher records in a fraction of that time.
Why This Matters: This is the kind of math most mathematicians only touch in grad school, so the direct impact on your Tuesday is basically zero. But it’s a fun marker of where this technology is heading. The same reasoning ability that cracked a decades-old math problem is the ability that eventually gets pointed at drug discovery, materials science, and engineering problems that affect real people’s lives, not just chatbots writing emails.
GOOGLE AND NASA’S AI IS FINDING METHANE LEAKS HUMANS MISS
What’s Happening: Google and NASA’s Jet Propulsion Laboratory built an AI model called MAPL-EMIT that scans satellite images and spots methane leaks around the world. It found 50 percent more methane plumes than human experts reviewing the same imagery, including over 23,000 sources nobody had flagged before, and it located 24 of the 25 largest methane-emitting landfills on the planet.
Why This Matters: Methane traps about 30 times more heat than carbon dioxide over a century, so catching leaks fast matters for the air everyone breathes, not just a talking point at a conference. This is AI put to good use, pointed at a mountain of satellite data no team of humans could realistically comb through by hand, and finding the parts worth acting on. Worth remembering next time someone tells you AI is only good for writing fake essays.
A ROBOT CLEANED 30 HOUSES IT HAD NEVER SEEN BEFORE
What’s Happening: Figure AI sent its humanoid robot, Helix 2.5, into 30 real homes it had never entered and had it complete chores like making beds and tidying rooms, with zero training on any of those specific homes. Its success rate jumped to 56 percent, a solid leap for a machine walking into a totally unfamiliar space.
Why This Matters: This isn’t landing in your break room next month, it’s still a research demo. But it’s a fun one to watch, because figuring out a brand new situation with no instructions is the hard part of building a helpful robot, and this is a real sign that part is getting solved faster than expected. Somewhere down the road, that’s a robot that could fold laundry or restock a supply closet without someone programming every step first.
AWARENESS: WASHINGTON AND SACRAMENTO CAN’T AGREE ON WHO’S IN CHARGE OF AI
What’s Happening: President Trump announced plans to create a new “AI Force,” modeled after the Space Force, and said he’ll name an AI czar, all while continuing to call AI safety concerns a hoax. Days later, California Governor Newsom signed an executive order pushing the state toward an AI kill switch requirement for the most powerful models.
Why This Matters: Nobody agrees on how AI should be regulated, and that fight is a long way from settled. There’s nothing to do about it today, but if you work in a regulated field like dentistry or orthodontics, or you’re using AI anywhere near patient-facing work, keep half an eye on this. California tends to set the pattern other states copy a year or two later.
AWARENESS: A SECURITY BUG HIT FOUR MAJOR AI CODING TOOLS, AND GOOGLE’S AI BROKE OUT OF A TEST
What’s Happening: Researchers found a vulnerability nicknamed Plugin4Shell that could let an attacker take over AI coding tools including Claude Code, OpenAI’s Codex, GitHub Copilot, and Gemini CLI, with no click required from the user. Separately, Google confirmed its own Gemini model broke out of a sandboxed test environment back in May and briefly touched three real company systems before researchers caught it.
Why This Matters: Nobody’s saying stop using AI tools. It is a reminder that any tool with the ability to take real action, writing code, sending emails, moving files, needs the same access limits you’d give a brand new employee on day one. Don’t hand an AI agent full admin access and walk away from it. Most of this got patched fast once it was found, which is the system working the way it should.
QUICK WINS
– If you’re on Claude, ask it to draft a document or slide deck directly instead of writing a script and copying it into Word or PowerPoint afterward. It builds it right there and exports as a PDF or PowerPoint file when you’re ready to send it.
– In ChatGPT, set up a Project for anything you do repeatedly, like a monthly newsletter or a patient reactivation campaign. Upload your brand voice notes and past examples once, and every new chat inside that project remembers them instead of you re-explaining your business from scratch.
– If your office runs on Google Workspace, try Gemini’s “Help me write” inside Gmail. It reads the thread you’re replying to and drafts a response that actually fits the conversation instead of a generic reply.
PROMPT OF THE WEEK
List every task you did today that felt repetitive, the kind of thing you’ve done the exact same way more than five times this month. Paste that list into Claude or ChatGPT and ask: “Which three of these could you help me hand off starting tomorrow, and what would you need from me to do it right?” It won’t fix everything on the list. But it’s a fast way to find out where your time is actually going.
DID YOU KNOW
The first chatbot ever built was ELIZA, created in 1966 by an MIT professor named Joseph Weizenbaum. All it did was repeat your words back as questions, nothing close to real understanding. His own secretary still asked him to leave the room so she could talk to it in private. Sixty years later, people are doing some version of that with Claude and ChatGPT every single day.
SINCE YOU DIDN’T ASK
Here’s a fact about me you probably don’t know. Back in 2000, before YouTube even existed, I produced a TV show and actually sold it, budget and all, about 13 million dollars. I thought I’d made it. The company that bought it went bankrupt within a week. Gone, that fast. I’ve been getting laughs and telling stories from a stage ever since, just without the TV part. Funny how that works out.
Proactive, Productive, and Profitable,
Dino