Keep up to date on the latest products, workflows, apps and models so that you can excel at your work. Curated by Duet.

Stay ahead with the most recent breakthroughs—here’s what’s new and making waves in AI-powered productivity:

Scriptly

Scriptly is a voice controlled teleprompter application for iOS designed to streamline video production for creators, solo entrepreneurs, and public speakers. Unlike conventional teleprompters that require manual scroll speeds or foot pedals, Scriptly uses on device speech recognition to match the scrolling text precisely to your natural speaking cadence. The app allows video creators to maintain steady eye contact with the camera while reading complex scripts effortlessly, making high quality video production faster and more authentic for social media and marketing channels.

Topview Motion Studio

Topview Motion Studio is an AI video generation workspace built for digital creators, ecommerce sellers, and small marketing agencies looking to produce promotional media quickly. By converting simple product links, media files, or landing pages into polished marketing videos, the platform eliminates the need for complex editing software or dedicated motion graphics teams. Specialized AI agents analyze current social media trends across platforms like TikTok to automatically generate relevant scripts, cinematic camera movements, and engaging visual layouts. This streamlined approach helps growing brands lower production costs while continuously launching fresh video ad variants to boost sales and consumer engagement.

Tucky

Tucky is a native macOS notes utility that transforms desktop workspace organization by docking subtly as a thin interactive bar on the edge of your screen. When hovered over, Tucky fans out into paper like tabs, enabling instant access to notes, tasks, and live Markdown documents without cluttering your active display. The application integrates a built in AI assistant that can answer questions, synthesize note contents, and trigger local workflows while keeping all underlying note data encrypted locally on your hard drive for maximum privacy.

AdAI

AdAI is an agentic marketing platform built to automate digital ad creation for small businesses and growth agencies. Rather than generating erratic visual content from loose prompts, AdAI utilizes specialized AI agents that reconstruct digital video and static image ads using a company's verified brand guidelines, approved media assets, and legal parameters. The platform connects with performance marketing data on platforms like Meta and Google to automatically tailor high converting ad variants in seconds.

Lightfield

Lightfield is an AI native customer relationship platform designed for growing startups and small sales teams seeking to eliminate administrative workload. The tool auto captures every email, calendar event, and video call, converting raw conversation streams into an organized database without requiring manual entry from reps. With its conversational assistant and automated workspace tracking, Lightfield surfaces pipeline insights, generates tailored follow up drafts, and keeps entire cross functional teams aligned around accurate customer data.

How Lightfield Is Reinventing Customer Management for Small Businesses

Managing customer relationships has long been one of the most tedious burdens for small businesses and growing teams. Sales representatives and founders spend countless hours each week manually typing meeting notes, updating deal stages, and logging email threads into traditional database systems. Lightfield changes this dynamic completely by introducing an AI native platform that updates itself automatically based on your actual everyday communication. By connecting directly to your email, calendar, and virtual meeting rooms, Lightfield captures every conversation and turns raw interaction into a structured, reliable system of record without requiring manual data entry.

The defining core of Lightfield is its automated world model engine, which synthesizes every customer touchpoint across your entire team into a single, unified intelligence layer. Rather than treating each email or call as an isolated event, the platform connects historical discussions, current action items, and shifting buyer sentiment into a comprehensive timeline. This allows every team member to step into any account with full historical context, drastically reducing ramp time for new hires and eliminating knowledge silos across sales, marketing, and customer support.

Beyond automated logging, the platform provides a conversational query feature that allows users to interact with their sales pipeline in natural language. Small business owners can ask detailed questions about deal bottlenecks, request instant summaries of past client discussions, or generate customized presentation drafts in seconds. The system actively analyzes client sentiment to flag deals at risk of stalling and suggests targeted follow up actions based on successful patterns observed in previous closed deals, giving small teams the tactical capabilities of an enterprise sales operations unit.

This shift from passive record keeping to proactive execution offers immense practical value for resource constrained teams. Instead of chasing updates or digging through buried communication channels, founders and sales leaders get instant, accurate visibility into pipeline health and deal velocity. By automating the administrative friction out of customer management, Lightfield gives teams back hours of valuable focus time every week, allowing them to concentrate entirely on building authentic relationships and closing deals.

The Answer Arrived. The Understanding Takes Longer.

Last week, OpenAI announced that a swarm of its AI agents had produced a proposed solution to a question about how fluids move — a question mathematicians had been unable to answer for roughly ninety years.

By the time the announcement went live, the fight over it was already public.

Not a polite fight, either. A mathematician had gone public hours earlier with an account of private calls involving pressure over authorship and a remark about his career. OpenAI disputes his reading of nearly all of it. Nobody outside those calls can settle who is right. And running underneath the whole mess is a quieter objection from one of the world’s leading mathematicians — written days before any of it happened — that probably matters more than the rest of it combined.

The question was about water

The Navier-Stokes equations date to the 1800s. They describe how fluids move — water in a pipe, air over a wing, blood in an artery — and engineers rely on them constantly.

They also come with a question nobody could close. The equations model a fluid as a continuous medium, and they’re written in terms of velocity fields that start out smooth. Does that smoothness always survive? Or can a fluid that begins calm wind itself up until some vanishingly small piece of it is moving infinitely fast — a “singularity”?

In 2000 the Clay Mathematics Institute gathered seven problems of this caliber, named them the Millennium Prize Problems, and attached a million dollars to each. Twenty-six years on, exactly one has moved into the solved column.

OpenAI’s answer is that smoothness does not always survive. In its construction, a three-dimensional fluid starts at rest, a carefully built external force is applied — “smooth” here is a mathematical term about how well-behaved the force is, not a claim that it’s gentle — and the velocity runs off to infinity in finite time while the total energy stays bounded.

None of this has any immediate practical consequence. Real water is molecules, not mathematical smoothness, so no river is going to tear a hole in itself. What it points to is a limit on how far the equations’ smooth description of fluid motion can be pushed.

Eighty-eight hours

Here’s how it happened, by OpenAI’s own account.

In the first days of this month, the company heard a rumor that a competitor was close to cracking a Millennium problem. Researchers pointed an unreleased internal model — one they say is considerably stronger than GPT-6 Astra, the flagship that went public while this was running — at the open problems on the list.

Agents worked in parallel, split into groups, each chasing a different angle, with results periodically pooled. The group that got there numbered around ten thousand. Eighty-eight hours after the first agents launched, they had a proof. Seventeen more hours and it had been rewritten in Lean, a language that mechanically checks every logical step of a formal argument — though specialists still have to confirm that what was formalized is actually the problem everyone means. Across all the problems attempted, the agents sent nearly five million messages to each other. Sébastien Bubeck, the OpenAI researcher who led the effort, put the computing cost at several million dollars.

Then OpenAI said it would not be claiming the million-dollar prize.

And here is where we are: A question open since the 1930s, closed in four days, by a project that only began because someone at the company caught wind of a competitor.

The two men in Madrid

Except this didn’t start from nothing.

For years Diego Córdoba in Madrid and Luis Martínez-Zoroa, who took his doctorate under Córdoba in 2023, had been building a way into this family of problems that almost nobody else was pursuing. Their approach stacks infinitely many well-behaved solutions into a cascade until a singularity falls out of the pile. By 2023 they had used it to prove blowup for forced three-dimensional Euler — the frictionless cousin of Navier-Stokes — but with a force that was mathematically rough. Getting that force down to smooth was the wall everyone then ran at.

Both teams that announced results this month were climbing that wall, on their road.

Charles Fefferman of Princeton, who wrote the Clay Institute’s official description of the problem, told Quanta that the heroes of the story are Córdoba and Martínez-Zoroa. Tristan Buckmaster, the NYU mathematician at the center of the other announcement, went further, writing that he believes Martínez-Zoroa deserves a Fields Medal.

Córdoba’s own line about his former student, as Quanta reported it, is the best thing anyone said last week: “I don’t use AI: I have Luis.”

Then it got ugly

Twelve hours before OpenAI published, Buckmaster and Levent Alpöge — a mathematician employed by Anthropic, though the two describe their collaboration as personal and unaffiliated — had posted results of their own, produced with heavy help from models built by both companies.

Theirs were not Navier-Stokes. They proved finite-time blowup with smooth forcing for three-dimensional Euler and two related systems. Real progress toward the Millennium problem; not the Millennium problem. OpenAI says its agents also resolved a version of Euler — the unforced one — and acknowledges the pair’s priority on the forced case.

What happened between the two camps is contested, and worth stating carefully.

In a public statement, Buckmaster says he learned on a call that OpenAI had gone after the same route he and Alpöge had quietly chosen — the path through a smooth force that Córdoba and Martínez-Zoroa opened. He says he asked whether the model had been trained on, or had access to, their Codex sessions, where they had parked every draft of the project, and got no answer on training. He says two proposals were floated: the pair could post first with OpenAI following a day later and publicly saying they deserved the Clay Prize; or Buckmaster could write up OpenAI’s Navier-Stokes result himself, crediting its model. He says Bubeck twice pushed for Alpöge to be left off that write-up because he works at Anthropic. He says he declined both, said he would go public, and was asked why he would ruin his career.

He is also explicit about what he isn’t saying. He hasn’t seen OpenAI’s proof, doesn’t know what its model did, doesn’t know whether their data was used, and states plainly that he is not accusing anyone — only recording what he was told.

Bubeck gives a different account. He says he never asked for Alpöge to be removed from the pair’s own work; the authorship question concerned a proposed rewrite of OpenAI’s proof, where he felt it improper for a rival lab’s employee to be an author. He says neither researchers nor agents saw the pair’s work before it was public, and that the two proofs differ substantially. On the career remark, he doesn’t deny it — he says he made it in frustration, retracted it on the call, and has apologized for the wording. Sam Altman backed him publicly within the hour.

One thread stays loose no matter whose account you prefer. OpenAI’s own writeup says no specific user data was accessed, then adds that it cannot rule out that de-identified data derived from the pair’s use of its products helped improve its models.

Notice what the fight is over. Not the money — OpenAI already waved that off. It’s over who gets to be the one who did it.

The objection that actually matters

Days before any of this broke, while the rumors were still only rumors, Terence Tao had already written the most unsettling thing anyone would say about it.

His target is specific: the fashionable practice of pointing a powerful AI at a problem, unguided by anyone with real expertise in the field the problem lives in. Used that way, he argues, producing answers and producing insight haven’t merely come apart — they’ve become negatively correlated. Optimizing harder for the answer can actively reduce what anyone learns from the problem, because the solution contaminates the reasoning, the signal-to-noise ratio degrades, and the incentive to keep studying the thing evaporates once it’s marked solved.

Tao is careful to say this tradeoff isn’t inevitable. He points out that using AI for literature search already produces genuine insight, and he’s praised AI-assisted mathematical work, including Buckmaster and Alpöge’s. What drives the wedge, he argues, is the incentive to race for priority above every other concern.

Which is worth setting beside everything described above. That race is not hypothetical. It is the entire reason this proof exists in the first place, on a timeline measured in days.

Tao’s other worry compares the practice to excavating an archaeological site with heavy machinery. You get the treasure and destroy the context that gave it meaning. Good open problems, he notes, are not a renewable resource — and if the search process and the failed attempts stay private, the dead ends that usually seed the next generation of techniques go with them. He floats declaring certain classes of problems off-limits to automated solvers, the way we’ve mostly agreed among ourselves not to spoil films.

The Clay Institute says something adjacent, right at the top of its page for this problem. Why insist on a proof at all? Because “a proof gives not only certitude, but also understanding.”

Last week produced a great deal of certitude. The understanding is still being worked out.

Why this lands on your desk too

Nobody is going to ask you to referee a fluid dynamics proof. But you will hear “AI solved it” for the rest of your working life, and this is a useful lesson in how much that sentence compresses.

Unpacked, this one involved: a ninety-year lineage of human work, a specific strategy built by two mathematicians in Madrid, a rumor that told the researchers where to aim, a formal verification system, several million dollars of compute, and a team steering the whole thing. What the machines did is real and unprecedented. They did not do it alone, and they did not do it from a standing start.

So when the claim lands in your industry — and it will — the three questions stay the same. Whose work was it actually built on? Who checked it? And is there understanding here, or only an answer?

That last one is the one people forget to ask. A good tool can make you faster and smarter both. But it will only make you smarter if somebody insists on understanding what came out the other end.

As of this writing, Clay’s website still lists Navier-Stokes among the unsolved problems; whether it recognizes this result remains to be determined. What’s certain is that the one prize ever awarded was turned down by Grigori Perelman, who thought Richard Hamilton deserved as much credit as he did. If a second is ever offered, it won’t be collected by a machine that can’t want credit at all.

Which leaves the wanting, as usual, entirely to us.

Partner Spotlight: Periscope VPN by Duet Display

Protecting your online privacy and securing your remote workflow should never feel complicated or slow. Periscope by Duet Display delivers a fast, seamless virtual private network experience designed specifically for modern professionals and small businesses on the move. Built with state of the art end to end encryption, Periscope keeps your sensitive data fully protected across public networks without sacrificing connection speed. The platform is remarkably easy to set up, requiring no complex configurations or technical expertise to get started. Best of all, Periscope operates under a strict commitment to user privacy with an absolute no logging policy, ensuring your online browsing habits, location data, and network traffic remain entirely confidential at all times. Periscope supports Mac, Windows, iPhones, iPads, and Apple TV.

Discover more and secure your connection today at getPeriscope.com.

Stay productive, stay curious—see you next week with more AI breakthroughs!