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:

Fambot

What if your family had an AI chief of staff? Fambot is a new consumer focused assistant designed to take the chaos out of school emails, newsletters, calendars, and family group chats. It turns all those scattered updates into a clear daily plan, surfacing deadlines, decisions, activities, and reminders so parents spend less time hunting through messages. Fambot launched this week with more than 1,000 families already using it and is currently free while in beta.

Causal

Planning a creative project can quickly become a mess of notes, images, links, sketches, and half formed ideas. Causal is a new AI powered canvas built to bring those pieces together in one visual workspace. You can ask its AI to organize ideas, brainstorm directions, build a roadmap, or create custom widgets directly on the canvas. It also supports connections to external AI agents, giving founders and creators a way to move from a visual idea board toward execution without constantly rebuilding the context somewhere else. Causal gaining attention among designers, creators, and founders.

Topview Motion Studio

Making a polished product launch video normally means learning motion design software or paying someone who already knows it. Topview Motion Studio takes a different approach. Give it a product brief, upload screenshots or product images, choose a visual direction, and it builds a connected launch video for you. Videos can run from 4 to 60 seconds and can be created in formats ranging from widescreen presentations to vertical social content. The tool is particularly interesting for small businesses and creators that need professional looking launch content without a professional motion design workflow.

Tidy

AI writing assistants usually require you to open another application or paste your text somewhere else. Tidy takes a much simpler approach. Select text in almost any Mac application, use a keyboard shortcut, and Tidy fixes spelling and grammar right where you wrote it. It also includes a second shortcut that removes the telltale style of AI generated writing while keeping your original meaning. Most interestingly, Tidy uses the Apple Intelligence model built into macOS, so the selected text stays on your computer rather than being uploaded to a remote AI service. It is free and requires macOS 26 with Apple Intelligence enabled.

Adobe for Slack

Adobe just put more than 70 of its creative and productivity tools directly inside Slack. With Adobe for Slack, you can describe what you want to create to Slackbot and have it call on tools including Firefly, Photoshop, Premiere, Acrobat, Illustrator and Lightroom. It can use the context already sitting in conversations, files and Slack Canvases to create images, videos and documents, then refine them through conversation. For a small marketing team, that means a campaign brief can move from conversation to finished creative without constantly switching between Slack and a collection of creative applications. The integration is available globally to Slack Business+ and Enterprise+ customers.

Periscope: Your Apple TV Just Became a Private Gateway Home

Most VPNs work by sending your traffic through a company's servers in a country you pick from a list. Periscope by Duet Display does the opposite. It turns a device you already own into a private gateway to your own home network, so your phone or laptop reaches the internet through your home connection instead of somebody else's data center. As of this month it runs on Apple TV, which means the server can be the box already sitting under your television.

The biggest improvement for everyday users is setup. Install Periscope on the Apple TV, choose Server on the first screen and pair your phone. There is no dashboard, no admin console, no port forwarding, no static IP address and nothing new to buy. If your bank app has ever locked you out while you were abroad, or a streaming service you pay for refused to play in the wrong country, that is the problem this solves. Your devices look like they are at home because your traffic really does leave from there.

That changes what a VPN is for. Instead of trusting a company you have never met with everything you do online, your traffic goes to your own house, and the only network in the middle is the one you already pay for. In other words, a VPN does not have to mean handing your browsing to a stranger in exchange for a promise of privacy. It can mean going home, which for a lot of people turns out to be the thing they actually wanted.

For consumers and small businesses, the practical benefit is simple: the files on the computer at home, the camera in the living room and the printer in the office all become reachable from anywhere. Periscope is on the App Store for iPhone, iPad and Apple TV, with Mac and Windows as a direct download at getPeriscope.com, and it needs one device at home that stays switched on. Do More readers can use the code DOMORE30 at check out for thirty days free. If purchasing from the app store use this link for the same thirty days free.

Do We Still Need Programmers?

There’s a version of this question that gets asked as a provocation, and a version that has quietly become one of the most consequential questions in the American economy. This is the second kind.

If a machine can write working code — and it can now, competently and at volume — what exactly is a programmer for?

The honest answer is that we still very much need programmers. What we appear to need a lot less of is the job that used to turn a person into one.

The number worth sitting with

Last month, three Stanford economists — Erik Brynjolfsson, Bharat Chandar, and Ruyu Chen — released a revised version of a paper they call “Canaries in the Coal Mine?” It runs on payroll records from ADP, the largest payroll processor in the country, covering millions of American workers through the middle of this year.

Their first finding is the one that gets skipped: there is no evidence of widespread, economy-wide job displacement from AI. The robots have not taken the jobs. Not broadly, not yet.

Their second finding is the one that should keep you reading. Employment among workers ages 22 to 25 in the occupations most exposed to AI now sits about 19% below where it would be if it had kept pace with similarly aged workers in less-exposed jobs. Experienced workers show no comparable gap at all.

That shortfall was 15% a year ago. It is 19% now. It has widened every time they have looked.

And it is showing up at the entrance, not the exit. The researchers are specific: the adjustment operates “primarily through reduced hiring of young workers rather than increased separations.” This isn’t a wave of firings. The door is simply not opening as often.

One clarification that matters, because a lot of coverage blurs it: this data sorts people by age, not by job title. Nobody in it is labeled “junior developer.” Ages 22 to 25 is a stand-in for early career, which is a reasonable stand-in and not the same thing.

Software developers sit squarely in the most-exposed group.

What AI actually took

The most interesting thing in the new revision isn’t a number. It’s a distinction.

The researchers separate two kinds of knowledge. Codified knowledge is the written-down kind — the standardized, documented, teachable material found in textbooks, manuals, and formal procedures. Tacit knowledge is the kind you can only get by doing: judgment accumulated through practice, mentorship, and repeated collisions with real situations.

Employment fell among young workers in occupations that lean on codified knowledge. Employment rose among experienced workers in occupations that lean on tacit knowledge.

The authors are careful here, and so should we be: they call this a descriptive exercise, not proof of a mechanism. And there’s a real wrinkle buried in their footnotes. The codified-knowledge pattern stops being statistically significant once you control for how college-educated an occupation is — the two are tangled together, and it’s genuinely hard to separate “AI ate the written-down work” from “something is happening to jobs that require degrees.” The tacit-knowledge result for experienced workers survives that same control.

So treat this as the most plausible story on offer rather than a settled finding. It happens to be a story that explains a great deal.

Generative AI is extraordinary at reproducing and recombining what has already been written down. It has been trained on a staggering volume of technical prose — tutorials, documentation, decades of Stack Overflow answers. What it cannot reliably do is the thing a fifteen-year veteran does without noticing: smell that a requirement is wrong before writing a line, know which part of the system will break in six months, understand that the customer asking for a faster horse actually needs a car.

Here is the trap. The entry-level job has always been the codified job. That was the entire design. You gave the new hire the well-specified, self-contained, documented tasks — precisely because they were the tasks that could be handed to someone with no judgment yet. Junior work was the tuition you paid to acquire the tacit knowledge that makes a senior.

If that reading is right, then AI didn’t replace programmers. It replaced the tuition.

Meanwhile, the profession is doing fine

Now for the part the doom coverage tends to leave out, because it complicates the story.

The Bureau of Labor Statistics currently projects employment of software developers to grow 10% between 2025 and 2035 — much faster than the average across all occupations, which is 3%. There were about 1.7 million software developer jobs last year, at a median wage of $135,980.

That is not a dying profession. That is a profession in the top tier of American earnings with a growth forecast most industries would envy.

But here the government makes a distinction worth stealing. BLS tracks a separate, narrower occupation it calls computer programmers — people who write, modify, and test code, largely turning designs made by others into working instructions. That category holds about 110,800 jobs, and BLS projects it to decline 7% by 2035. The agency’s stated reason is worth quoting: many companies are expected to use technologies “including artificial intelligence (AI), to automate repetitive programming tasks,” while higher-skilled work shifts to developers.

So the literal answer to the question in the title, as far as the federal government is concerned, is: fewer of them. The people who translate a specification into code are shrinking. The people who decide what the specification should be are growing.

Both things are true at once, and holding them together is the entire skill of thinking clearly about this: the field is expanding while its entrance is narrowing. The people already inside are doing well. The people trying to get in are facing a door that has been getting heavier for nearly four years.

A caution about that 19%

The Stanford authors are more careful with their own finding than most of the people quoting it, which is worth respecting.

They call these results descriptive, not causal — “canaries in the coal mine” rather than proof. The estimated gap shrinks, and in some specifications stops being significant, once you account for education. Some of the divergence was visible before generative AI showed up. The gap looks larger in the ADP data than in national survey benchmarks. Their own strongest claim is only that the results are consistent with AI having begun to affect entry-level employment.

They also address the obvious alternative explanation. Plenty of people argue this is simply the hangover from cheap money and pandemic-era overhiring — that companies stopped paying juniors to learn when interest rates made it expensive, and blamed AI afterward because it sounded better. That argument has real force. But the divergence has continued to widen well after rates peaked, it persists when you exclude tech firms and computer occupations entirely, and it clusters specifically in jobs where AI is used to automate rather than assist. Interest rates alone don’t predict that pattern.

And there is real evidence pointing the other way. Economists at the New York Fed looked at job postings earlier this year and found something awkward for the AI story: demand for AI-exposed occupations had already been falling before ChatGPT, with no clear new break afterward. More pointedly, they found no clear gap between junior and senior postings inside highly exposed occupations — the two moved roughly in parallel. Their conclusion was that AI “may be contributing,” but is not the main driver.

Two serious teams, two data sources, two different answers. Payroll records say young workers are falling behind; job listings say the slowdown isn’t specific to them. That disagreement is the actual state of the evidence, and anyone telling you this is settled — in either direction — is selling something.

The machines are not as good as the panic suggests

There’s a second half to this that rarely makes the headlines.

In mid-2025, the research nonprofit METR ran an unusually rigorous experiment: a randomized controlled trial with experienced open-source developers working in codebases they knew intimately. The result was startling — with AI tools, they took 19% longer. Worse, they couldn’t tell. Afterward, the same developers estimated AI had sped them up by 20%.

That study got enormous mileage. It is also now obsolete, and METR says so plainly at the top of the page.

What happened next is the better story. That first trial ran on the tools of early 2025. By the time METR tried to repeat it, developers had moved on to agentic assistants like Claude Code and Codex — software that doesn’t just suggest a line but goes off and does the task. The follow-up gathered far more data: 57 developers, more than 800 tasks. Then the researchers threw out their own estimate and went back to redesign the study.

The reason is the interesting part. Developers kept declining to take part, because they didn’t want half their work assigned to a no-AI condition. Among those who did sign up, 30% to 50% said they had chosen not to submit certain tasks for the same reason — no misconduct, just people quietly protecting their own week. One compared working without it to walking across the city after getting used to taking an Uber. METR is honest that a pay cut, from $150 an hour to $50, probably contributed too.

Be careful what you take from that. It is strong evidence that developers now place real value on these tools. It is not a measurement of how much faster the tools make them — that number is exactly what got thrown out.

And even so, developers don’t trust the output. In Stack Overflow’s most recent developer survey, 84% were using or planning to use AI tools — while the share trusting the accuracy of what those tools produce fell to 33%, down from 43% the year before. Nearly half now actively distrust it. Adoption up, confidence down. The most-cited complaint, from 66% of respondents, was code that is almost right. Not wrong enough to fail loudly. Just wrong enough to cost you an afternoon.

Catching “almost right” requires knowing what right looks like. That is tacit knowledge again.

The irony nobody planned

Stack Overflow’s own guidance to developers on working with these tools, published earlier this year, recommends thinking of an AI coding assistant as a junior developer: fast and promising, but prone to basic errors and in need of supervision and redirection.

It’s good advice. It also contains the whole problem in one sentence.

The industry has decided to treat AI like a junior developer at the precise moment it sharply reduced junior hiring. Which raises a question nobody has answered: if the entry-level rung is where seniors come from, and that rung is narrowing, where do the supervisors come from in 2036?

Every company making this trade is making it individually rational and collectively reckless. Skipping the junior hire saves money this quarter. It also quietly consumes a shared resource — the pipeline of people who will one day be senior enough to catch what the machine got almost right — that no single company has any incentive to refill.

This was never only about code

Programmers are just the clearest case, because software is one of the most thoroughly documented crafts in existence and therefore among the most legible to a machine trained on documents.

But look at the shape of the thing and you’ll find it everywhere. The first-year paralegal doing document review. The junior analyst building the model. The entry-level copywriter drafting the product descriptions. The résumé screener, the bookkeeper, the translator. Any field with an apprenticeship rung where the newest person does the well-documented work may face a version of the same squeeze — and the Stanford data show these effects concentrated exactly where AI substitutes for human tasks rather than assisting with them.

If you are early in a career, the strategic implication is uncomfortable but clear: the codified part of your job is no longer where your value lives. Learn the fundamentals anyway — you cannot supervise what you don’t understand, and you cannot catch “almost right” without knowing right. But get to the tacit work as fast as you possibly can. Sit in on the client call. Ask why the decision went the other way. Take the ambiguous problem nobody has written the manual for.

If you manage people, the question is whether you are still willing to pay for someone to learn. That cost used to be invisible, absorbed inside a junior salary. It is now a line item you have to choose on purpose.

So: do we still need programmers?

Yes — though not quite in the shape the question assumes. We need fewer people to translate a specification into working code, and more people who can judge whether the specification was any good. Both trends are already in the federal projections, pointing in opposite directions.

What we have not worked out is how anyone gets from the first job to the second. We have started treating the making of a programmer as someone else’s expense — asking machines to do the work that used to turn beginners into experts, while still expecting a steady supply of experts.

The canary is not dead. It’s just that the youngest bird in the mine has stopped singing, and everyone has agreed to describe that as a productivity gain.

Partner Spotlight: AutoDoc Meeting Transcription by Duet Display

Meetings create a mountain of information, but most meeting assistants stop at a transcript or summary. AutoDoc goes further. It is a free, open source AI meeting assistant that runs locally on your computer and can continue operating even when you are disconnected from the Internet. That means your private information never needs to leave your computer.

AutoDoc records both audio and video, and your notes are synchronized with the video. So instead of simply reading what happened during a meeting, you can go back and see exactly what was presented at the moment a topic was discussed. AutoDoc is available as an open source project for developers and privacy minded users, and there is also an installer available for anyone who simply wants to download it and start using it.

Find out more information or sign up at getAutoDoc.com.

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