rendoire

Streaming Was Practice

Sat, 25 Apr 2026 00:16:53 GMT

How we lost intentionality to algorithms, and why AI is the same trick with higher stakes.

On October 23rd 2001, Steve Jobs held up a white rectangle and said 1,000 songs in your pocket. That line aged better than most things from that keynote. The iPod held 5GB. Songs were 3 to 5MB. You got a thousand slots. The ceiling forced you to decide.

Look at your Spotify home screen. How many of those playlists did you actually make?

Twenty-five years later you have access to a hundred million songs and most people cannot name the last album they chose. They can quote their Wrapped. They can show you the playlist they did not make.

Tim Wu calls this pattern The Master Switch. Every information industry follows the same cycle. Open, centralised, monopolised, disrupted, open again. Telephone. Radio. Film. Television. The internet. It happens every time. Streaming was the last cycle. AI is the next one.

The iPod was a philosophy, not a product

A caveat first. For nearly eighty years before the iPod, the dominant mode of music consumption was radio. Someone in a tower somewhere decided what played. The iPod was not the default. It was a brief anomaly in a century of feeds.

Which is why it mattered.

Getting music onto an iPod was a ritual. You heard a song somewhere. You wrote it down. You went to iTunes and paid 99 cents, or you opened LimeWire and waited twelve minutes for a file that was a virus half the time. You tagged the ID3. You dragged it into a folder. You synced.

Then, because you only had a thousand slots, you deleted something.

Every add was a commit. Every delete was a commit. Your iPod was a curated self. Hand someone your iPod and in ten minutes of scrolling they knew you. Not because of what an algorithm thought. Because of what you chose.

Jay-Z put it plainly on Diamonds from Sierra Leone:

“I’m not a businessman, I’m a business, man.”

Owning your library made you the business. The iPod was the output. The iTunes library was the asset. The library was sovereign. You could copy it, back it up, take it off the grid, play it in a field with no signal. The music you owned would outlive the companies that sold it to you.

That is gone. Most people under 25 have never owned a song.

Napster was the bridge, not the enemy

The conventional story is that Napster killed the record industry. The honest story is that Napster taught the industry what it would become.

Napster launched in June 1999. By early 2001, 80 million people had used it. Metallica sued in April 2000. They were not wrong. They were early.

Napster did not just make music free. It normalised infinite access. Before Napster, scarcity was the default. After Napster, the only question was bandwidth. The library became the cloud, before anyone called it that.

The record industry fought piracy for a decade and lost. Then in 2008, Daniel Ek and Martin Lorentzon pitched the labels a deal: what if the thing Napster did, infinite access, was legal, and we gave you a cut? The labels, bleeding out, agreed.

Piracy did not lose. Piracy was legalised and sold back to us on a subscription.

The file disappeared.

You do not have a Spotify library. You have an account. When Neil Young pulled his catalogue over Joe Rogan in 2022, albums vanished from users’ “libraries” overnight. When Taylor Swift re-recorded her early albums because Scooter Braun owned the originals, the streaming versions got quietly demoted. You rented access to a catalogue that can be reorganised while you sleep.

The algorithm did not replace the radio.

Streaming did not replace the iPod. It replaced the radio. Then it dressed up the radio as a library, with your face on it.

The first generation of streaming still asked you to choose. You searched. You built playlists. You were still the DJ.

Then Spotify shipped Discover Weekly in 2015. Daily Mix. Release Radar. AI DJ in 2023. Each one took a little more of the decision. By the time TikTok’s For You Page took over attention around 2018, you were not choosing. You were reacting.

The architecture removed the forcing function of decision. You no longer had to commit to a thousand songs. You had all songs. Functionally, you had none. A library you did not build is not a library. It is a feed.

The standard defense is that the algorithm exposes you to things you would not find alone. Yes, and. Engagement-optimised variance is not editorial taste. Blogs, IRC, Pitchfork before it calcified, college radio, mixtapes from friends introduced more genuine novelty than any For You Page, because they optimised for interesting, not sticky. The algorithm does expose you to new things. It just narrows what new is allowed to mean.

If you want it in black and white, Meta publishes it. Their transparency page for Instagram Reels Chaining, updated November 2025, lists what the system is actually optimising for. Not is this reel good. Not will you remember this in an hour. The first prediction on the list is literally how likely you are to watch less than three seconds of a reel. The rest of the stack is the same shape: how likely you are to reshare, to use the audio in your own post, to follow the author, to click Interested, to watch longer than 95% of users who saw a reel of the same length. The system is not trying to help you find things worth your attention. It is modelling your attention as a probability distribution and optimising for continuation. A Pitchfork editor was trying to tell you what was interesting. Instagram is trying to predict whether you will click off in the next three seconds. Those are not the same goal and they do not produce the same output.

This is not nostalgia, modern streaming has better ergonomics in every measurable way. What got worse is your agency over your own taste. You consume more content than anyone in history and you can recall less of it.

You are what you repeat

“You are what you eat” is the lazy version. The serious version is that cognition is built from inputs. What you attend to, repeatedly, rewires you.

Daniel Kahneman’s Thinking, Fast and Slow gives us the vocabulary. System 1 is fast, intuitive, pattern-matching, mostly unconscious. System 2 is slow, deliberate. The part of you that actually chooses. System 2 is where identity gets made. It is also lazy by default. It will let System 1 run the show unless you force it awake.

The algorithm is a System 2 bypass.

You do not decide what to hear next, because the feed decides. You do not evaluate, because the feed pre-evaluates. Every scroll, every auto-play, every recommendation trades a moment of cognition for a moment of ease. Stack enough of those and System 2 atrophies.

The algorithm’s job is to convince you otherwise. Sometimes it is right. The problem is that better is measured by engagement, not by whether it made you sharper.

Now the same story, faster

Streaming ate your taste. If the algorithm picks your music, the worst case is you become culturally boring. Bad, but survivable.

AI is coming for your reasoning. Outsourcing your taste makes you a passive consumer. Outsourcing your reasoning makes you redundant. Those are not the same stakes. It is the why in whatever it is you’re doing.

You used to learn by experience + refinement. You had a coding problem. You spent four hours on Stack Overflow. You wrote something bad, then something better. The friction was the feature. The frustration was the mechanism by which the circuit that could solve problems like this actually formed.

Previously LLMs were not at a utility to consumers, outside of researchers and universities in 2018. Then OpenAI shipped ChatGPT in November 2022. Four years later, the default move for any cognitive task is prompt first, think later. Not think and prompt. Prompt, and think only if the prompt fails.

I am not anti-AI. I run small models on my own hardware. I build agents, utilise AI Workflows and architecture. The capability is real. The question is what the use is doing to you.

The obvious objection is the calculator. Nobody thinks we got dumber when we stopped doing long division. The calculator replaced arithmetic and we moved up a layer. Why is AI different?

Two reasons.

Arithmetic has a verifiable answer. Synthesis does not. When you outsource the first draft of a thought, you are not editing toward truth. You are editing toward fluency. Fluency is what the model is trained to produce.

You build that model by writing bad drafts and fixing them. Skip the drafts and you are not editing. You are rubber-stamping a machine that optimises for confidence and encouragement to the user.

The real question is not the model. It is the stack beneath it.

For most of 2022 to 2024, the dominant narrative was scale wins. GPT-4, Claude 3, Gemini Ultra. The frontier was a club of four labs. Peter Thiel’s Zero to One had the script: monopoly is good, plan for monopoly not competition.

Then it broke.

Meta released Llama 2 with open weights in July 2023. Mistral’s 7B landed that September and punched above its weight. Microsoft’s Phi series kept shrinking while performance held. By 2024, a 14B model on a consumer GPU could do what a 175B model did in 2022. In January 2025, DeepSeek’s R1 (open weights, a fraction of the training cost) made the point unmissable. The moat was thinner than the incumbents claimed.

The distinction that actually matters is not small vs large. It is portable vs rented.

A rented/frontier model (GPT-5, Claude via API, Gemini) lives in someone else’s datacentre. You send a prompt, they send a response. If they turn it off, retrain it, raise the price, or decide your use case violates a new policy, you are done. Your workflows break. You are building on rented land.

A portable model (MiniMax, Mistral, Qwen, DeepSeek) runs on hardware you control. Slower. Less capable on the hardest benchmarks. Will not match a frontier model on cutting-edge reasoning today. Currently this with Google’s Gemma4 2B & 4B parameter models are crushing it & with innovations to processing chips, KVM & efficiencies in context, this is only going to get better.

This is not a purity argument. Keep a rented frontier model for the genuinely hard problems. Claude and GPT-5 are better than any 8B at the edge of the capability curve, and that matters sometimes. What it does not mean is that the frontier should be your default. For 80% of knowledge work, a local model handles it and the difference does not touch the output. Portable by default. Rented for the edge cases.

The distinction compounds when you zoom out.

Agentic AI. An agent reads your email, files your tickets, ships your PRs. Every action is a moment of trust. Do you want the agent that runs your life to be a rented process subject to policy changes you do not see? Or running locally, auditable, with weights you can control. Take the plug out & control your data.

Robotics and action models. Google DeepMind’s RT-2, NVIDIA’s Project GR00T, the broader vision-language-action push. These are AIs that move things in the physical world. A humanoid robot that cannot act without a datacentre roundtrip is a paperweight. On-device models are not a niche. They are the foundation of physical AI.

Enterprise. Every serious enterprise in the last year ends the same way. Frontier models win demos. Fine-tuned open models in your VPC win production. Data sovereignty, compliance, cost. All push the same direction.

Entrepreneurship. If you are building on a frontier API you are one pricing change away from your margins vanishing. Every solo dev who built on GPT-3 in 2021 learned this when the terms shifted. The businesses that survive own their stack.

Donella Meadows’ Thinking in Systems names the move. Leverage points. You do not change a system by fighting its outputs. You change it by finding the structural lever where a small shift produces a disproportionate effect. The leverage point here is not use less AI. It is own the stack beneath the AI.

The Graveyard of Convenience

Before you decide convenience is worth it, walk through the graveyard.

Google Reader died in 2013 and took a generation’s RSS habits with it. Evernote’s free tier got paywalled into uselessness. Twitter shut its API in 2023 and killed an ecosystem of third-party clients overnight. Parler got deplatformed at the infrastructure layer and lost its users’ data with no recourse. Every indie dev who built on GPT-3 in 2021 learned the hard way that pricing and terms shift under live products. Zapier, Notion, Figma. Every SaaS you depend on has a pricing roadmap you do not see.

If you do not host it, you do not own it. If you do not own it, you are renting. If you are renting, the landlord can evict you.

The Technical Maturity Model

You do not need a Kubernetes cluster to start. Three tiers. Stop at any of them.

Tier 1. Digital Hygiene. Five minutes.

Open ChatGPT, Claude, and Gemini settings right now. Turn off use my data to train future models. Almost nobody does it. It takes less time than a coffee order. Audit your home screen. Delete one app whose only job is to decide for you. For one task this week, try it yourself for two minutes before you prompt.

Tier 2. Local Inference. An afternoon.

Install LM Studio or Ollama on your laptop. Click to install. Pick a small model like Llama 3.2 or Phi-4. It will not match Claude on the hardest problems. It will handle 80% of your daily prompting and never send a byte to anyone else. Replace one cloud service with a local one. Bitwarden instead of 1Password. Obsidian instead of Notion.

Tier 3. Full-Stack Reclamation. A weekend.

This is the tier where you stop being a consumer of infrastructure and become its operator.

Repurpose an old laptop into a home server. A used HP ProDesk costs less than a month of cloud subscriptions. Install Proxmox. Run Ollama. Expose it to your devices over Tailscale. Now you have a private inference endpoint in your office. No one can turn it off. No one can raise the price. No one can retrain it without telling you.

Self-host the things you use. Jellyfin for media. Nextcloud for files. n8n for the workflows you would have built on Zapier. Vaultwarden for passwords. Each dependency reduced is a piece of your cognitive life that cannot be repossessed.

Stewart Brand said it at the 1984 Hackers Conference:

“Information wants to be free.”

The full quote, less often cited, continues: “Information also wants to be expensive. That tension will not go away.” We are living in the expensive half. The way out is the same as it was in 1984. Build your own tools. Own your own data. Be suspicious of anyone who tells you the convenient thing is also the free thing.

The best technology is human

Fri, 03 Apr 2026 21:25:11 GMT

What was the first technology that made you wow, it likely wasn’t the app you use to check your bank. When asking yourself this question, you are likely to have a moment where you felt, i want that.

That was how I felt about both the iPod Classic & iMac. That translucent, colour that looked nothing like the beige boxes cluttering every office? It wasn’t the fastest computer you could buy. It wasn’t the most powerful. It couldn’t even run half the software that “serious” computers could. But something about it made you want it.

That’s the thing about technology’s greatest moments — they’ve never really been about the technology at all. They’ve been about how that technology made us feel, how it fit into our lives, how it solved problems we didn’t even know we had. Yet somehow, we’ve forgotten this lesson. We’re drowning in a sea of benchmark scores and feature lists, competing over marginal improvements that most of us can’t even notice.

And now, as AI reshapes everything around us, we’re at a crossroads. Do we repeat the mistakes of modern tech, building systems that impress investors more than they help humans? Or do we remember what made technology magic for anyone from a coder to a musician.

When Design Beat Specs (And Why We Loved It)

1984: The Mac That Smiled Back

It’s 1984, and computers are these intimidating, beige monuments to productivity. IBM PCs dominate offices with their serious black screens and cryptic commands. Then Apple rolls out the Macintosh and it has a face. A little smiling computer that seemed to say “hello”.

Steve Jobs wasn’t trying to build the fastest computer. He wanted to build one that felt like a companion(R2-D2). People didn’t buy Macs because they crunched numbers faster, they bought them because using one felt good.

Attention vs. Intention: How We Lost the Plot

Sun, 04 Jan 2026 03:43:16 GMT

Look at the apps on your home screen right now. Ask yourself if this tool is waiting for you to use it or if it is begging you to look at it.

There has been a massive shift in how technology is built. We went from an era of Intention, where tech was a bicycle for the mind, to an era of Attention, where your focus is a natural resource to be mined and sold.

Here is how we got here, why every app feels the same, and why the “shiny object” diet is killing actual innovation.

“The computer is the most remarkable tool that we’ve ever come up with. It’s the equivalent of a bicycle for our minds.” — Steve Jobs (1990)

1. The Era of Intention (When Tech Was Boring but Useful)

In the beginning, computers didn’t care if you liked them.

The early days from the mainframes to the 90s were about utility. You used a computer to crunch numbers, write a paper, or organise a database. You didn’t hang out in a spreadsheet just for fun. You had a specific goal, you used the tool to achieve it, and then you walked away.

Even the counter-culture boom of the 70s and 80s was driven by a desire to empower people. This was the era of the Homebrew Computer Club and the garage days of Steve Jobs. It was hippie idealism meets soldering irons.

The vibe was simple. Here is a superpower you didn’t have before. Go build something.

2. The Shift: When You Became the Product

Somewhere along the line, the business model changed. Companies had to figure out how to pay the bills if they gave software away for free.

They borrowed the answer from TV. Ads.

By the mid-2000s, platforms like Facebook realised that a user who logs in, does a task, and leaves is worthless. But a user who stays for 4 hours is a goldmine.

The metric for success switched from User Utility to Time on Device.

The Result: We went from tools that respect your time to casinos that steal it.
The Stat: The average attention span has dropped to about 8 seconds, which is less than a goldfish, and we check our phones roughly 96 times a day.

We aren’t using the tools anymore. The tools are using us.

3. Why Everything Looks the Same

Have you noticed that every app is morphing into the same thing?

Instagram used to be for photos, but now it is TikTok. YouTube used to be for long videos, but now it is TikTok Shorts. LinkedIn is somehow also becoming Facebook.

Every app is just a vertical feed with an algorithm designed to keep you scrolling.

We have processors in our pockets that would have been supercomputers in the 90s. We could be using that power to solve energy crises or cure diseases. Instead, we use that silicon to render high-def ads and filter our faces in real-time.

There is no risk anymore. In the Dot Com boom, people tried weird stuff. The iPod changed music. Online gaming created new worlds. Now we just get a slightly faster iPhone every September and a chatbot that regurgitates data.

It is a monopoly problem too. Apple and Samsung basically own the hardware. Amazon and Microsoft own the servers. When a few giants own the board, nobody wants to flip the table.

4. The Shiny Object Trap

The biggest loss isn’t our attention span. It is the next generation of builders.

Right now, the smartest 22-year-old engineers are being trained to think that innovation means keeping people addicted to a screen.

If we feed young inventors a diet of shiny objects and dopamine hits, they won’t grow up wanting to solve hard problems like logistics, healthcare, or education. They will grow up wanting to build a better ad algorithm.

We are teaching them that Marketing Tricks are more important than Real Solutions.

The Conclusion: Let the Pendulum Swing

We are overdue for a correction.

The 90s tech counter-culture thought the internet would free us from corporate control. Instead, it just built different cages. But the feeling is starting to shift again. You can see it in the rise of dumb phones, the privacy regulations in Europe, and the general fatigue with being online 24/7.

To get back to an era of Intention, we need to stop measuring success by how long we stare at a screen. We need to measure it by what we achieve when we put the phone down.

Real innovation isn’t a better feed, automating everything or a chat bot. Real innovation is building a tool so good that you can finish your work and go outside.

Think Different: The Dyslexic Advantage in Systems Thinking

Tue, 16 Sep 2025 23:25:50 GMT

Apple told us to “Think Different.” For dyslexic engineers, thinking differently isn’t a choice. It’s how their brains are wired. And in complex systems work, that difference is becoming a superpower.

Apple’s old campaign had it right: “Think Different.” Albert Einstein, Thomas Edison, Richard Branson, Steve Jobs. All dyslexic. All systems thinkers who changed the world.

What if thinking differently isn’t just a marketing slogan but an actual cognitive advantage?

The Pattern Spotters

Working in software systems, I’ve noticed something curious. When it breaks, you may find that one engineer who spots the pattern everyone else misses. An issue in service A connects to latency in service B, which links to database exhaustion from that deployment three days ago.

That’s not luck, that’s a different intelligence.

The Science Behind It

Research from Cambridge University is rewriting what we know about dyslexia. Dr. Helen Taylor’s team found that dyslexic brains aren’t broken. They are specialised to explore the unknown. Optimised for pattern recognition across complex systems.

Brain imaging shows these advantages come from fundamental neural differences present from birth.

Why Complex Systems Need Different Thinking

Traditional analysis breaks problems into parts. Dyslexic cognition naturally grasps how parts interact as wholes. Researchers call it “global exploratory learning.” Understanding complexity through interconnections rather than components.

Dyslexic engineers often excel at:

Spotting system-wide patterns in data
Designing resilient architectures
Reducing Noise & Improving Signal rate
Connecting seemingly unrelated incidents

The Entrepreneur Connection

A 2003 study found that 40% of successful entrepreneurs are dyslexic. That’s four times the general population rate.

Richard Branson: “My dyslexia became my massive advantage. It helped me think creatively and laterally, and see solutions where others saw problems.”

Entrepreneurship is systems design. Identifying market gaps, orchestrating resources, creating value networks. The same cognitive traits that make reading difficult make systems thinking natural.

Building Better Teams

The best engineering teams combine different cognitive styles. Detail-oriented linear thinkers ensure correct implementation. Big-picture systems thinkers ensure architectural coherence.

Cognitive diversity reduces single points of failure in problem-solving.

Changing the Story

We’ve spent decades focused on what dyslexic people can’t do. Reading difficulties. Spelling problems. Processing delays.

Researchers now advocate for “shifting focus to dyslexic advantages.” In technology, this matters. Instead of seeing dyslexia as something to overcome, we can recognise it as cognitive specialisation suited to complex systems work.

The Hidden Architecture

Every complex system has hidden architecture. Patterns and dependencies that determine how it behaves under stress. Understanding this requires the kind of global, pattern-oriented thinking that dyslexic cognition provides.

The engineers who excel at this level often share certain traits: they see connections others miss, think in visual terms, focus on interactions rather than components.

These aren’t accidents. They’re enhanced abilities vital for adapting to changing environments.

Why This Matters Now

As systems become increasingly complex (distributed cloud, AI systems, global supply chains), dyslexic thinking becomes more valuable, not less.

The question isn’t whether dyslexic individuals can succeed in business & technology. The evidence shows they do. The question is whether organisations will leverage it.

Learning of my own dyslexia not only liberated me from comparison but enabled me to work in a way that matters to me.

Think Different

References:

Developmental dyslexia essential to human adaptive success, study argues

https://onlinelibrary.wiley.com/doi/10.1002/dys.388

Reclaiming Your Digital Autonomy: Practical Self-Hosting in an Age of Data Harvesti and…

Wed, 30 Jul 2025 10:00:48 GMT

Reclaiming Your Digital Autonomy: Practical Self-Hosting in an Age of Data Gathering and Gamification

Our daily lives are increasingly shaped by digital services that promise convenience at the hidden cost of our privacy and autonomy. Each notification, like, and tap is carefully engineered to capture our attention, turning everyday interactions into a relentless pursuit of engagement. Cal Newport, author of Digital Minimalism, reveals how digital platforms exploit psychological vulnerabilities to monopolise attention and transform users into data sources:

“You cannot expect an app that was built to hijack your attention to help you reclaim your focus.”

Similarly, Cory Doctorow, in The Internet Con, describes how platforms initially offer value, only to degrade the user experience once they achieve market dominance — ultimately prioritising data extraction over user satisfaction.

AI and the Intensification of Surveillance Capitalism

The rapid evolution of artificial intelligence has dramatically intensified these dynamics. AI systems rely heavily on vast datasets and immense computational resources, favoring large corporations that control the necessary infrastructure. Melanie Mitchell, in AI: A Guide for Thinking Humans, points out that AI’s current limitations compel companies to double down on data collection to sustain performance.

Shoshana Zuboff, in The Age of Surveillance Capitalism, underscores the threat:

“Surveillance capitalism unilaterally claims private human experience as free raw material for translation into behavioural data.”

Modern AI tools often act as mere interfaces for centralised platforms like OpenAI, consolidating data control in the hands of a few.

Self-Hosting: A Path to Digital Sovereignty

Self-hosting offers a powerful way to counteract centralisation and reclaim digital control. By hosting services yourself, you manage your own digital infrastructure, reducing reliance on corporate platforms and shielding yourself from surveillance and gamification.

Practical Steps for Digital Self-Reliance

Here are actionable steps you can take to begin self-hosting:

Personal Cloud Storage: Replace services like iCloud or Dropbox with a home NAS system such as Synology or TrueNAS.
Secure Communication: Transition to self-hosted messaging and email platforms like Matrix, Rocket.Chat, or Mail-in-a-Box to avoid corporate surveillance.
Media and Entertainment: Host media libraries with tools like Jellyfin or Plex, offering privacy-focused alternatives to Netflix and Spotify.
Ad and Tracker Blocking: Use Pi-hole or AdGuard Home to block intrusive ads and trackers across your network.
Local AI and ChatGPT Alternatives: Set up your own AI chat interface using OpenWebUI paired with open-source LLMs for private, local AI interaction.
Automation and Agentic AI: Use tools like n8n to automate tasks privately, creating AI agents that operate under your control.

Replace Big Tech Tools

Opt for self-hosted alternatives:

Nextcloud for document collaboration
Bitwarden for password management
Home Assistant for smart home control

Repurposing Old Hardware

Repurpose older desktops or laptops as home servers to cut costs and reduce e-waste. These machines can efficiently run essential services such as storage, media streaming, and even AI models.

Consumer Electronics and AI Accessibility

As consumer hardware evolves, it can either empower or limit access to independent AI capabilities. Frameworks like Mixture of Experts (MoE) will eventually enable high-performance AI with significantly lower computational demands, allowing individuals to achieve near-corporate-grade AI performance without needing massive infrastructure, thus democratising access.

Advantages

The mixture of Expert (MoE) architecture offers several advantages over traditional deep learning models.

Enhanced scalability: MoE models scale easily to billions or even trillions of parameters due to sparse activation, which reduces the need for massive computational power.
Increased flexibility: One unique benefit of MoE is that new experts can be added to the existing model without retraining the entire system. This adaptability allows the model to readily accommodate new tasks and domains.
Efficiency: Since MoE activates only the most relevant experts for each input, it can handle diverse tasks more efficiently than traditional models. This makes it faster and more accurate, as experts can focus on what they do best.
Parallel processing: Experts can work independently, which allows for efficient parallel processing. This approach can cause faster training and inference times.

The Hidden Costs of Convenience

Thomas Thwaites, in The Toaster Project, reveals how seemingly simple products depend on vast, hidden layers of complexity and extraction. Digital services are no different; beneath their ease of use lie serious costs: privacy erosion, data commodification, and diminished autonomy.

Samuel Florman, in The Existential Pleasures of Engineering, reminds us:

“Technology is not what alienates us; it is what connects our intentions with the physical world.”

Digital Minimalism Meets Self-Hosting

Combining digital minimalism with self-hosting creates a resilient framework for reclaiming autonomy. By rejecting superficial digital engagement and embracing intentional, privacy-respecting habits, you foster genuine independence.

Reclaiming your digital sovereignty through self-hosting is more than a technical decision - it’s an act of resistance, restoring autonomy and dignity in an era of increasing surveillance and algorithmic manipulation.

As Cory Doctorow encourages:

“Build the world you want,” rather than accept the one handed to you.

Built Different: The Structural Shift of Agentic AI

Wed, 02 Jul 2025 22:53:00 GMT

“The real problem is not whether machines think but whether men do.” — B.F. Skinner

We’ve built software that listens. We’ve built software that learns. But now, for the first time, we’re building software that does — without waiting for us to tell it how.

Evolution of Internet Capable Technologies

This isn’t just a shift in capability. It’s a shift in structure. Agentic AI is the first mainstream paradigm where outcomes are emergent, not explicitly programmed. And like any system that’s both powerful and opaque, its benefits will not be distributed equally — and its risks won’t be obvious until after something breaks.

1. The Myth of Smarter Tools

Adding more agents to your business won’t fix your inefficiencies — it’ll expose them. Brooks’ Law (from The Mythical Man) reminds us: adding manpower to a late software project makes it later. The same applies to agents. Complexity scales faster than control.

Agentic AI doesn’t eliminate the need for design. It intensifies it. You’re now coordinating semi-autonomous tools with their own memory, feedback, and behavior. And unless you design for drift, you will get it.

The Fast and the Furious: Tokyo Drift (2006)

Most builders will learn the hard way: your first agentic workflow will fail — because systems without constraints grow unpredictable.

2. From Apps to Agents: A Systems-Level Shift

Apps are reactive. Agents are systemic.

Apps wait for input. You tell them what to do.
Agents infer, plan, act — and replan. They loop.

“You’re not building tools anymore. You’re designing behavior.”

What does this mean?

In systems thinking (as introduced by Donella Meadows), a loop is a structure that creates feedback. Agents live inside loops — and they create them. Every observation changes what they plan. Every action changes the state of the system.

Agentic AI Logic Flow

Unless monitored, these feedback loops introduce instability. This is the beginning of non-deterministic software in mainstream use — and that demands systemic literacy, not just programming skill.

3. Cognitive Load Is Shifting, Not Disappearing

Delegation moves effort up the stack.

You spend less time doing the thing — but more time specifying what the thing is, how it should happen, and what to do when it doesn’t.

This is what systems thinkers call model externalization. You can no longer rely on implicit judgment. You must expose your assumptions and expectations — and structure them into observable boundaries.

The mental work didn’t vanish. It just became structural.

4. The Unfair Advantage Will Be System Design

Agentic workflows introduce an entirely new failure mode: the right output for the wrong reasons.

You didn’t ask for X, but the agent gave you a convincing version of it — rooted in a flawed assumption or a poorly scoped boundary. Now you’re running an operational risk with an illusion of competence.

This means: — Design prompts like you write software specs — Audit behavior, not just results — Create fallback systems and observability hooks

Think DevOps + Product + Psychology. Your team’s edge will not be model access. It’ll be system clarity.

5. Small Builders Face a Wall They Don’t Know Exists Yet

In 2008, the App Store let anyone ship. That era is gone.

Steve Jobs at WWDC (2008)

Agentic AI is built on top of infrastructure that’s expensive, proprietary, and fast-moving. Fine-tuning a foundation model isn’t something you do in your bedroom.

That means the power to shape behavior is centralizing. If we’re not careful, the age of AI will be less open than the age of apps.

6. How to Prepare If You’re Not a Lab or a Billionaire

In the short term:

Only build agents for workflows you deeply understand
Use manual review steps and circuit breakers
Monitor not just what agents do — but why they did it

In the medium term:

Learn how systems fail, not just how they succeed
Treat your prompt logic as code — version it, test it, trace it
Read systems theory — not just machine learning blogs

In the long term:

Build internal libraries of task specs and safety scopes
Design for correction speed, not just first-try performance
Focus on understanding behaviors across time, not just in snapshot outputs

Final Thought: This Is About Power

Every architecture distributes power.

The web distributed access. Apps distributed utility. Agentic AI will distribute agency — the capacity to act on intent.

But only if you own the design, the scaffolding, and the feedback loops.

You’re not competing on intelligence anymore. You’re competing on delegation clarity and systemic foresight.

Build accordingly — or be built around.

I Used a Dumb Phone for a Month — With a Modern Twist

Wed, 02 Jul 2025 00:03:51 GMT

📵 I Used a Dumb Phone for a Month — With a Modern Twist

“Clutter is costly, and its cost is often measured in terms of reduced autonomy.”

— Cal Newport, Digital Minimalism

As a software engineer, technology is central to everything I do. It’s my tool, my craft, and the foundation of how I build and create. But there’s a fine line between using technology intentionally — and being consumed by it.

The infinite scroll, the dopamine-driven app design, the subtle but constant pull to check, swipe, refresh — it wasn’t just a distraction. It was rewiring how I thought and how I spent my time.

I didn’t want to abandon tech — but I did want to reclaim control.

So I made a deliberate switch: I replaced my modern smartphone with a “dumb phone.” Not to disconnect entirely, but to reset my relationship with digital tools and reintroduce friction where it mattered.

In a world engineered for distraction, I decided to reclaim my focus — not by throwing away all tech, but by changing how I used it. For 30 days, I switched to the Unihertz Titan Pocket, a smartphone in a dumbphone’s body. The goal wasn’t full disconnection — I still needed Spotify, navigation, and banking — but I wanted to break the endless scroll loop.

Here’s what I used, how I set it up, the benefits, drawbacks, and whether I’d recommend it.

🧠 What Even Is a Dumb Phone?

A dumb phone typically refers to:

No app store (or heavily limited)
No social media
No touchscreen or modern UI
Just calls, texts, and maybe Snake

But there’s a spectrum:

Classic Feature Phones

Nokia 3310, Alcatel Go Flip

Fully off-grid or minimalist users

Minimalist Smart Devices

Light Phone II, Punkt MP02

Designed for peace of mind

Locked-Down Smartphones

Android/iOS with launchers or focus apps

Reduce distraction while keeping power

Hybrid Form Factors

Unihertz Titan Pocket

Retro design, modern software control

🔧 Why I Chose the Unihertz Titan Pocket?

Unlike a true dumb phone, the Titan Pocket gave me:

Contactless payments
Banking apps
Android Auto for navigation
Spotify & YouTube (audio only)
Niagara Launcher for a minimal, swipe-up app interface

Its small screen and QWERTY keyboard limited mindless usage. I didn’t install video-first apps like Instagram or TikTok. If I ever did get the urge to scroll, I opened daily.dev — a feed of insightful articles, not dopamine-chasing junk.

Evenings? Just like 2007.

When I needed to check social messages, I visited the web versions of platforms. This alone helped distance me from the addictive nature of app-based interaction.

✅ The Benefits

1. More Presence

I was there — whether it was golfing with friends or just sitting with someone. No mental pull to “just check something.”

2. Less App Fatigue

No pop-ups. No forced updates. No trending nonsense I didn’t ask for.

3. Better Use of Time

By design, the device discouraged passive use. I was intentional. I’d listen to music — not scroll. I’d read an article — not react to a reel.

4. Awareness of the Invisible Transaction

Modern apps aren’t just tools. They extract attention — and sell it. Using this phone made that painfully obvious.

⚠️ The Drawbacks

1. Security Risks

The Titan Pocket runs Android 11, and support has ended. I used Bitdefender, but long-term OS vulnerabilities remain a concern.

2. Form Factor Friction

The small screen made scanning gym barcodes annoying. Some UI elements on apps weren’t optimized for this retro form.

3. Cloud Syncing Issues

My photos were still locked into iCloud. Since the Titan runs Android, accessing or syncing those wasn’t seamless. (I’m solving this in a future post about switching to NAS for storage.)

💭 Final Reflection: Would I Recommend It?

Yes — for anyone who wants a reset without sacrificing modern essentials.

The Titan Pocket gave me:

Modern utility
Old-school discipline
A clearer head

You don’t need to go full caveman. But you do need to be hyperaware of how today’s devices are designed — not for your benefit, but for your attention.

📌 TL;DR

📱 Used: Unihertz Titan Pocket
🧠 Why: Reduce distraction, retain essentials
✅ Pro: Focus, intention, peace
❌ Con: Security, screen size, app compatibility
👍 Recommendation: Yes, especially if you’re feeling digitally burned out

Would you try something like this? Or have questions about setting one up yourself? I’m thinking of writing a guide on creating your own “distraction-free device” — let me know.