Innovation: Weekly Summary (November 17-23, 2025)

Some weeks are noisy. This week felt like that — lots of little bright things and a few heavy ones. I would describe them as a jumble of front-page announcements, deep think pieces, and folks trying to make sense of where technology actually helps people. To me, it feels like a map being redrawn while people argue about which roads to keep and which to tear up. I’d say the conversation kept circling a few familiar spots: agents and models, tinkering versus coordination, the politics of innovation, and the everyday business of turning ideas into products. Below, I walk through what I noticed, note the tensions, and point to the posts that are worth a longer read if you want the original take.

Platforms, protocols, and where text lives now

There was an old‑school, web‑native vibe in a post by daveverse. He launched a new app that stitches his blog to WordPress and turns it into ActivityPub content for Mastodon. He muses about Bluesky and its limits. The headline idea is simple: text still matters. Not the shiny reels or the short bursts you scroll past, but the plain block of words that you can link, quote, and think about later.

I’d say this is the sort of nostalgia that also doubles as a practical bet. To me, it feels like someone cleaning out a drawer and finding a reliable tool. You don’t toss it because it’s old. You put it back in daily use. There’s a clear argument here: the oldest web pieces — linked text, readable posts — still solve real problems. They are portable. They survive platform churn.

If you’re the sort who likes small rebellions against design trends, this will ring true. The post hints at a bigger debate: do we rebuild social systems around text, or accept that the shiny format wins attention? It nudges you toward the first choice, but it does that by showing a working patch, not a manifesto.

AI agents, models, and the fight over how they should behave

This week had a lot of noise about agents and big models. There are at least two camps arguing loudly.

Nate (nate) dug into Google vs Vercel vs Anthropic and tried to cut through the documentation pages. The practical nugget is this: agents are not just fancy chatbots. They are attempts to automate multi‑step workflows. Nate breaks down where each company bets — which workflows they target, how they handle context, and how they try to reduce the daily grind.

Then Google went big with Gemini 3, covered by Brian Fagioli. The rollout included multimodal capabilities and a new Gemini Agent for multi‑step tasks. The message was: more context, more reasoning, and a push into working with the user instead of replacing them. A related note: Google also released Nano Banana Pro for image generation, which one writer called a potential 'ChatGPT moment' for images. That’s the kind of product push that redefines expectations overnight.

Meanwhile, posts like Agents are Strategic Software by Peteris Erins emphasize that adoption is shallow in many places. Lots of companies value AI as a shiny metric, but few actually scale it into their operations. His point is blunt: if you want real benefit, focus on growth and innovation, not just cutting costs. He also introduces the phrase "agentic workflows" — that is, workflows where software can run small experiments and make strategic pivots — and says that’s the real future.

These pieces together make a small argument that keeps popping up in my head: the technology is moving fast, but the playbook for sensible adoption hasn’t caught up. There’s the technology on one side and actual work practices on the other. It’s like buying a new pressure cooker and expecting your grandma’s stew to taste better immediately. You need time to learn the knobs.

A related practical post offered a custom prompt for figuring out careers with AI agents. Small detail, but useful. It felt like a pocket tool. If you’re curious about where to use agents in your daily job, that piece nudges you to look at repetitive, multi‑step tasks. The advice is surprisingly tactical.

Failure, tinkering, and the messy path to useful things

A theme that kept coming back was the role of failure and messy experimentation. There were several posts that pushed against tidy narratives of single‑hero breakthroughs.

One post titled "Failure Is Required" by Otakar G. Hubschmann makes a point with a friendly grit: failure is not shameful, it is the feedback loop. He invokes WD‑40’s many tries and suggests that generative AI product building especially must embrace failure. The tone was encouraging. It reminded me of those kitchen experiments when a recipe goes wrong and you end up inventing something better. I’d say the message is that if your team avoids small failures, it will miss big wins.

That dovetails with a gently iconoclastic piece — the one that rejects the "great man" idea of innovation. It suggested innovation is more like cooking than heroics: you throw ingredients in, you taste, you tweak. Sometimes you get lucky. Sometimes you make a mess. But you keep trying. The writing had a jokey streak, mentioning bizarre inventions and how odd successes can come from very odd places. That small chaos is important. It is not glamorous, but it is where progress usually hides.

Then there’s a longer discussion about "deep tinkering vs deep coordination" from Drawn In Perspective. It felt like putting two cousins side by side. On one hand, tinkering thrives when feedback is quick and cheap. Think of hobbyists building radios in garages. On the other hand, coordination matters as projects scale. Once you need many people or big capital, the cost and friction of coordination start to throttle innovation. The post lists three barriers: when to hand off tinkering to organized teams, making tinkering accessible, and making the handoff effective when it happens. Those are nice categories. They make you imagine a relay race where the baton is fragile.

There’s mild tension between the celebration of tinkering and warnings about coordination. They’re not enemies. They’re like coffee and sleep — you need both at different times. The posts make the point that our institutions and funding models need to accept that truth.

Public money, private bets, and who should pay for the risk

Government money showed up too, with a solid case that public funding still builds foundational technologies. Naked Capitalism ran an analysis arguing that publicly funded patents are few but disproportionately powerful for GDP growth. The headline claim: 2% of patents are publicly funded, but they account for about 20% of medium‑term productivity gains. That’s a big ratio. The post warns that cuts to agencies like the NIH and NSF are not just budget politics. They are slow leaks in the innovation ship.

Jason Crawford’s piece — "Il Progresso in Crisi" — also echoed a worry about complacency. He traced the idea of progress and argued that belief in progress and investment are not automatic. I’d say his tone was both historian and cheerleader with a caveat. He wants people to see progress as fragile. It’s a reminder that innovation needs a culture and funding to keep it alive.

Then there’s the immigration debate. Bryan Caplan wrote two pieces that touch this seam. One argued opponents of low‑skilled immigration ignore how cheap labor can actually encourage innovation in some ways. The other piece critiqued J.D. Vance’s framing that cheap labor is an addiction. Caplan’s pushback was, roughly: dependencies aren’t bad if they are useful. His writing suggests we shouldn’t romanticize the idea that removing cheap labor will magically force better technology. It’s a useful check against certain policy fantasies.

Put these together and you get a picture: public funding, private entrepreneurship, immigration flows, and cultural belief in progress all form a tangled policy knot. Pull one thread and the whole pattern shifts. Policy talk here felt less ideological and more practical. People were arguing percentages, not slogans.

Domain experts and the next wave of startups

One piece that stood out was Todd Gagne’s "The Domain Expert Revolution." It argues that the next big startups will come from people who have deep industry experience, not just from generic tech founders. The examples were concrete — like wildland firefighters building tools to coordinate resources. To me, it feels like a practical correction to the old Silicon Valley myth of the outsider starting something huge from a dorm room.

The post says domain experts see inefficiencies others miss. They also come with trust and context. The trick is pairing them with strong technical execution. That’s the glue. If you take this advice, you end up with more realistic products that actually serve users, not shiny demos.

There’s a local spin too. David Cummings wrote about local startup hubs and AgeTech. He reminds us that innovation is often a neighborhood thing. Hubs matter. They create place‑based serendipity. If you’ve ever been to a local co‑working space and seen a random conversation lead to a partnership, you know what he means. These hubs are messy but they work — like a farmer’s market where the best ideas sometimes hide between the apple stalls.

The legal tech summit and the problem of scale

The TLTF Summit in Austin showed how a tight community can scale awkwardly. Robert Ambrogi covered this. The event is invitation‑only and avoids sales pitches. People liked the food, the vibe, and the candid talks. But there’s a common event problem: as attendance grows, the very thing that made it special — small‑group serendipity — gets harder to keep.

That’s a small human problem. It’s like your favorite pub getting crowded. You still get a pint, but the chat you relied on becomes harder. The takeaway is subtle: community‑first events are valuable, but they must guard the parts that made them useful. The legal tech world, where trust matters and practice is specialized, needs those intimate spaces to actually operationalize AI in their workflows.

Engineers, technical decisions, and the poker table of code

Kyle Cascade used a poker metaphor to talk about engineers deciding when to hold, fold, or walk away from tech. It’s an appealing frame. Many decisions about rewriting or keeping legacy systems are contextual. Chesterton’s Fence keeps popping up — don’t remove structures you don’t understand. The author also mentions the Innovator’s Dilemma and how AI might extend the life of certain stacks rather than force rewrites.

I liked the hands‑on vibe. This is not Silicon Valley armchair theory. It’s practical. It’s the day‑to‑day decisions that shape what we can actually build. If you work with code, you’ll nod and grimace along.

Hardware, manufacturing, and the small shifts that feel big

Apple’s move to 3D‑printed titanium for the Apple Watch drew attention from Jonny Evans. The company is using 100% recycled aerospace‑grade titanium powder and a fancy printing process. The claim is twofold: better products and less waste. That’s both an engineering and a PR play. It’s also one of those small, incremental manufacturing changes that, over time, matter a lot.

Imagine swapping one part of your car for a version that wastes less metal and lasts longer. It’s not headline‑shaking in the moment, but it ripples through supply chains and expectations. Apple likes to do that. It sets a bar and forces suppliers to respond.

On a different hardware note, Ian Mansfield wrote about seeing IBM’s quantum computer in London. People love the drama of quantum. The post made it clear that the machine is more of an exhibit than a consumer device. You can see it behind glass. It’s a signpost: the tech exists, but it’s still in a special room.

Drones, the Dandelion Tank, and the strange school of wartime innovation

War is a merciless accelerator of tech. That came through in two pieces by David Axe and a piece on drone survival in Ukraine. Small, cheap drones and electronic warfare have reshaped modern combat. One article notes many drones last only two minutes in contested airspace — that’s astonishing and brutal. The technology evolves fast because lives and territories are at stake.

Russia’s "dandelion tank" is a creative, if grim, response: a tree‑like structure to catch or deflect drones. It’s a vivid image. It’s also a reminder that innovation isn’t always about apps or labs. Sometimes it’s a physical contraption you bolt on because the battlefield demands it. These posts carried a chilling, practical tone: innovation there is direct. There’s no focus group, just immediate adaptation.

Methods, science, and the continuity of odd practices

The "Methods" post argued against a one‑size‑fits‑all scientific method. It suggested that real science looks more like a family resemblance than a recipe. That’s an interesting nudge for innovation practice too. If science uses many methods, then so should innovators. Try, test, model, argue, and sometimes just do the weird experiment that doesn’t fit your grant box.

This ties back to tinkering again. Methods plural means you keep many tools in the shed. That’s important because innovation rarely fits tidy checklists.

Small teams, looped feedback, and the people side of creativity

Rands in Repose published "The Loop," a piece about creativity and teamwork. It argued the creative spark needs clear expectations, a bit of discomfort, and strong feedback loops. In plain terms: give people tasks, but don’t smother them. Push them to do more than expected. Accept mistakes. The post read like advice from someone who’s seen teams make or break products.

I’d say this is the human glue of innovation. You can have the best model or the fanciest machine, but if the team doesn’t get clear signals fast, the project stalls. That’s the recurring theme: feedback loops matter.

What keeps appearing, like a pattern on a well‑worn shirt

Several threads kept showing up in different corners. First: the gap between shiny tech and adoption. You see a new model, an exciting hardware trick, or a clever agent. Then you read the adoption pieces: it’s harder than it looks. People need workflows, funding, and cultural space to absorb the change.

Second: the balance between tinkering and coordination. Early innovation loves cheap experiments. Scaling needs paperwork, governance, and sometimes boredom. That friction is the place where a lot of projects fail, or morph into something else.

Third: the politics and funding of progress. Public research matter, immigration flows alter labor economics, and private firms make bets. All of these influence what gets built and who benefits. The conversations this week pushed back against simple claims. They asked for nuance, and for decisions grounded in real numbers or real practices.

Fourth: domain experts are back in vogue. The idea that deep field knowledge plus execution wins — that theme kept repeating in startup and product posts. If you want durable startups, look for people who know the problem like the back of their hand.

Fifth: failure is not a dirty word. Multiple writers suggested we need more small, safe failures to find the winners. That felt both practical and humane.

There’s a small human detail I enjoyed: the constant return to analogies. Whether it was cooking, a poker table, a farmer’s market, or a pressure cooker, the writers used everyday images to explain big ideas. It makes these posts less like academic papers and more like conversations you’d have over coffee.

If you want to go deeper, follow the links in these notes. Each post hides practical tactics or historical digs that deserve a closer look. The week didn’t resolve the big questions. But it did give a decent pulse check: innovation right now is noisy, contested, and alive in many small places. It’s not a single thing. It’s many small things, some messy, some lovely, all worth a look.