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

I would describe this week in software blogs as a bunch of people talking to each other across small fences. Some of them shout about big shifts. Others fuss over the tiny things that nibble at our patience every day. It feels like standing at a busy crossroads. You can smell the coffee from the corner shop, hear a bus sigh, and notice two people arguing about whether to fix or to replace the old engine.

The shape of the shift: verifiability and "Software 2.0"

Andrej Karpathy throws a big idea into the pond and watches the ripples. He talks about AI not as a small feature but as a whole new way of computing. To me, it feels like someone saying that we moved from horses to cars — not just faster travel, but a different kind of trip. Karpathy calls it "Software 2.0". I’d say that phrase nails the mood. He argues that the tasks AI handles best are the ones you can check after the fact. If you can verify the outcome, you can automate it. If you can’t verify it, it stays stubbornly human.

That verifiability bit is interesting because it explains why some things took off and others didn't. It’s like folding laundry: if someone can look and say "that shirt is folded right," you can teach a machine to fold. But if the judgment is subtle, like whether a poem feels sincere, you’re out of luck. Karpathy ties this back to history. He lines up AI next to past computing paradigms. It reads like a neat timeline. But it also nudges a practical point: product teams should ask, "Can we check this easily?" If yes, maybe hand it to AI. If no, keep a human in the loop.

That idea of verifiability came back later in other posts, quietly. When people complain about annoying updates or missing features, they are often saying the same thing: the product's designers didn't measure what matters to the user. So Karpathy’s post is the big-picture frame. It gives a reason why some software feels modern and why some still feels stuck.

Agents and strategy: software that thinks like a shop manager

Peteris Erins is thinking more like a business owner. His post is less dreamy and more tactical. He looks at how companies actually adopt AI. The headline is blunt: valuations are sky-high, but only a few firms scale AI in ways that matter. He wants teams to focus on growth, not just on squeezing costs.

His key idea is "agentic workflows" — software systems that act more like assistants who can make strategic moves, not just execute scripts. To me, it feels like going from a microwave that heats to a sous-chef who knows when to add salt. He argues these agentic pieces let a company pivot faster. That matters. The old playbook — automate boring tasks to save money — is small-minded, he says. Better to build workflows that help you discover new wins, that hunt for opportunities.

Erins also lists common pitfalls. He’s pragmatic. He warns about organizations trying to bolt on agents without changing how decisions get made. It's that classic thing: you don't just buy a power tool and expect your garage band to sound like a stadium tour. If you give a team a new toy but keep the same approvals and meetings, the toy sits under the table.

There’s a link back to Karpathy here too. If you want agents to act right, you need metrics and verifiability. Otherwise they might wander off like a stray dog.

Old-school internet nostalgia, and a new feed dance

daveverse — Dave Winer, who pretty much wrote parts of the web early on — is quietly violent about feeds. He writes about FeedLand, which is not just another feed reader. It's a system that uses dynamic OPML lists. That is slightly geeky-sounding, but the point is simple: make feeds more flexible and shareable.

Dave is excited about networks of feed apps. He keeps bringing up the RSS era. He remembers a time 22 years ago when feeds were new and they changed things. I’d say he’s trying to restart that energy. The image I had reading this was of a local farmers’ market. Everyone brings something. Some people swap recipes. New apps could do that kind of sharing again.

He even mentions the possibility of linking with other products like Overcast or Pocket Casts. It’s not a manifesto of war on big platforms. It’s more like someone saying, "We can make a neat neighborhood. Want to help?" There’s a soft invitation in his tone. If you miss the early web's smell of possibility, Dave’s note will tug you.

The software that’s needy: updates, notifications, and the user who just wants calm

Michael J. Tsai nails a very different corner of the conversation. He writes about how many apps feel needy. You know the feeling: your phone buzzes, your computer nags you about updates, and some app insists you go through onboarding again. It’s that friend who keeps rearranging your furniture.

Tsai’s post is a slow burn of frustration. He calls out mandatory updates and developer priorities that ignore user control. He points at big names — Adobe, Apple — and says, "You're doing this to yourselves." The complaint is specific. It’s not a general moan. He lists examples and points out that updates often serve the company more than the user.

This is where Karpathy’s verifiability idea haunts the conversation. If an app kept features verifiable and predictable, maybe it wouldn’t need to bother the user as often. Tsai wants software that respects the user’s attention and time. Imagine a neighbor who fixes things once every blue moon rather than drilling at dawn every Tuesday. That’s what he’s asking for.

I’d say his piece is the one that will make you check your notification settings immediately. It’s practical. And a little righteous.

Life without Windows: small liberation stories

dorinlazar.ro writes in a different register. Their piece, in Romanian, tells the tale of three months using Linux — Fedora 42. It reads like a diary. They left Windows because the Microsoft experience had become a fight. After three months, they say they barely need Windows at all. Occasional use for a few apps, sure. But mostly, Linux covers what they need.

What’s worth noting is the tone of surprise and calm. Games? Mostly fine. KDE? Feels a lot like Windows. Console? Optional. That last bit matters because some people assume using Linux means living in a terminal. This author keeps saying, "Nope, it’s okay. You can do this." It’s like someone saying you can choose to ride a bicycle instead of a car and not die in the rain.

This piece connects to the needy-software theme. The author left Windows partly because the updates and product decisions felt like a betrayal. Linux, for them, feels like stepping into a quieter house. Not perfect — nothing is — but a better fit for their life.

Small app updates that matter: Calibre and the comfort of useful fixes

Brian Fagioli reports on Calibre 8.15. This one is the kind of post that rewards people who care about small things. The update adds a simple but useful tweak: highlights now show date tooltips. That’s the kind of detail that, if you use it, you will love. It’s the dusting of a bookshelf. Preserving formatting in comment edits, new keyboard shortcuts, stability fixes — these aren’t flashy. But they count.

What I liked about this post is that it treats routine improvements as worth noticing. People who maintain digital libraries get a lot from small quality-of-life changes. The news system adding sources is a nice touch too. It makes Calibre feel less like a silo and more like a little community tool.

There’s something comforting about this kind of attention. Not everything has to be revolutionary. The world also needs folks who sweep the floors and fix the squeaky hinge.

DearMob: an alternative approach to iPhone backups

Amerpie by Lou Plummer wrote twice this week. One piece is a hands-on look at DearMob iPhone Manager. This post reads like advice from someone who has lost data and learned to be cautious. The author prefers DearMob over Apple’s built-in backup because it offers selective control, speed, and lots of file support.

He’s pragmatic about the downsides. Third-party tools can break when Apple changes things. But he also notes the current price is attractive — a Black Friday-style nudge. The review is aimed at both "normies" and power users. I’d describe the tone as friendly and practical. He’s saying, "If you care about control and don’t want to fight iCloud, try this."

This ties back to the needy-software thread. Backups are where you feel the consequences of bad software decisions. DearMob is pitched as a way to take back control. It’s like buying a safe for your papers. Maybe you don’t need it, but when something goes wrong, you’ll be glad it’s there.

Obsidian: love for linking, patience for rough edges

Relja Novović offers a thorough review of Obsidian and its Sync service. The piece reads like someone showing you their notebook and then pointing out the stubborn smudge on the corner of the paper. The reviewer praises Obsidian’s internal linking and the free model. Those links are the product’s heart. They let you stitch notes together in a way that feels alive.

But Relja doesn’t gloss over problems. Filename sanitization is awkward. The Zettelkasten workflow isn't perfect out of the box. Still, the overall feel is: this tool helps you think better if you're willing to learn its quirks. There are tips and small configuration notes that will help newcomers. The review is practical and earnest. It’s like a friend telling you, "You’ll like this, but don’t be surprised if you have to tighten a few screws."

Obsidian’s story loops back to Karpathy’s idea. Tools that let you connect things — and then check the results — are more likely to become part of how people work. Obsidian gives you the map and the string. It doesn't always tell you where to pin the labels.

Black Friday wish lists and the impulse to collect apps

Back to Amerpie by Lou Plummer again, the Black Friday wish list is a lighter, more personal note. It’s the kind of list you make when you like gadgets and also like a good discount. The author confesses to collecting software. The items range from CSV Table to Digital Photo Frame to Anybox and FileMinutes.

This piece is useful in a small way. If you’ve ever felt tempted by a sale, you’ll see yourself in it. It’s human. It’s the same urge that fills a closet: a little fear that if you don’t buy now, you’ll miss something. The author’s tone is frank about the habit. There's a wink in the writing — like, "Yeah, I have too many apps. But these are sweet deals." It’s comforting, oddly.

The wish list also quietly points to a broader habit: people buy tools hoping to solve a problem later. Sometimes it works. Sometimes you end up with a drawer full of apps like jars in a pantry you never open.

Recurring themes: control, verification, and attention

If you step back, a few ideas keep bumping into each other.

• Control vs automation. Many posts are about giving control back to users. DearMob, Obsidian, Linux — these are all about choosing where control lives. Then you have Karpathy and Erins saying some control can be handed over to AI, but only if you can check it.

• Verifiability and metrics. Karpathy’s strong view that automation likes verifiable tasks shows up indirectly in the other posts. When things break or when updates feel arbitrary, it’s often because the company isn’t measuring the right outcomes.

• Attention and software’s neediness. Tsai’s complaints are echoed in small ways elsewhere. Even a good tool can become needy. The fix isn’t always new features. Sometimes it’s fewer interruptions.

• Tools as neighbors. Dave's feed vision and Obsidian’s internal linking both push toward small networks of tools that talk to each other. There’s a longing for community-style, interoperable tools rather than one giant platform that decides everything for you.

• Small wins matter. Brian’s Calibre update and Relja’s tips show that not all progress is headline news. Tiny changes add up.

Tensions and disagreements

Not everyone is singing the same tune. Karpathy’s embrace of AI as a new paradigm can sound like a promise of quick fixes, while Tsai’s and dorinlazar.ro’s writing prefers steadiness and control. Erins tries to bridge that gap by saying: use agents, but use them for growth, not only savings. In other words, don’t use AI to shave pennies if you could use it to find new lanes.

Dave’s nostalgia for open feeds sits opposite the realities that many users now live with: closed ecosystems that are easy and comfy. Open systems ask for a little more elbow grease. That trade-off shows up again in the Linux post. You can choose freedom and a slightly higher learning curve, or choose convenience and let someone else manage things for you.

Amerpie’s pieces highlight the economic angle. Deals and niche utilities are part of the ecosystem. They don’t settle the big philosophical debates, but they’re the practical choices people actually make.

A few practical nudges I picked up reading all of this

• If you build software, ask how you will verify it. Don’t just build a feature and hope users get it. Measure meaningful outcomes.
• If your company is adopting AI, think strategy first. What new things can AI let you do, not just what costs it can cut?
• If your app nags users, maybe rework the attention model. People will pay with their time or they will leave. That’s a choice.
• If you’re roaming the market for tools this Black Friday, think about whether you really need the tool or if it’s a pantry jar you won’t open.
• If you’re curious about leaving Windows, read firsthand accounts from people who made the jump. They’re not shouting. They’re practical and a little relieved.

These are small nudges, not commandments. They’re the kind of things you might argue over in a café. You might take one, tweak it, or toss it. That’s fine.

Little asides and tangents (because conversation rarely runs straight)

Sometimes a tech blog reads like a letter from an uncle who repaired radios in the seventies. You get a mix of big ideas and shop-floor advice. That’s what I found here. The feeds post made me think of swapping homemade jam at a farmer’s market. The Linux diary made me picture someone finally getting a coat that fits. The Obsidian review felt like a friend letting you borrow a book and then warning you not to fold the pages.

Also, the phrase "agentic workflows" tickles the imagination. It conjures an office full of polite, efficient clerks who also happen to be tiny wolves in suits, finding opportunities. That’s a goofy metaphor, but it stuck.

One more thing: reading all these posts in a row is like listening to a city for an afternoon. You hear the construction, the old man whistling, the barista scolding a colleague, and then a busker playing a tune that makes your shoulders drop. Some posts are the busker. Some are the jackhammer. Both matter.

If you want more detail on any of these threads, the authors linked above go deeper. They have the examples, the screenshots, the command-line notes, and the little annoyances that make these essays worth a proper read. Follow the links. You’ll learn different things from each post. Some will surprise you. Some will make you nod and mutter, "Yep, exactly."