Blog

Introduction

As 2025 draws to a close, I wanted to share some insights on what I think will be important in the coming year.

With that in mind, I want to discuss my thoughts on several key areas:

• The RAM Crisis
• Agentic AI and Workflows
• Invisible AR
• AI Scams
• The Blue Collar Boom

Let me dive into each of these insights in detail.

The RAM Crisis

I predict that the current “artificial” RAM shortages will intensify over the next few months unless big tech companies face government regulation to prevent monopolistic behavior and maintain fair access to memory resources.

Since 2022, the semiconductor industry has already been struggling with capacity limits for High Bandwidth Memory (HBM) and DRAM, driven by the explosive demand from data centers and AI accelerators. Production ramps for newer 3nm nodes are still catching up, creating a “tight-rope” scenario where supply barely meets the voracious demand of AI models.

That being said, I think Google completing its vertical alignment with AI is a massive advantage. Unlike its competitors, Google relies on TPUs (Tensor Processing Units) rather than the standard NVIDIA GPUs. Their approach is proving to be more efficient and sustainable for long-term AI development, which allows them to sidestep some of the supply chain bottlenecks we are seeing elsewhere.

Of course, Nvidia, Microsoft, OpenAI, and Anthropic are not happy with Google’s position. Currently, Google is the only major player not fully participating in the “cannibalistic” bidding wars for hardware that are inflating share prices across the industry.

I believe there is a bubble—similar to what Michael Burry predicts—but it may not reach the catastrophic levels of the 2008 crash. Unlike 2008, where every bank was exposed, this bubble is concentrated among a few key tech players. While valuations of AI-related chip makers have surged, the broader market dynamics will determine if the correction is severe or just a market adjustment.

If the bubble pops, the market will undoubtedly be in disarray for a bit. I am not certain if diversified giants like Microsoft will be heavily damaged, but I am certain that AI companies without full vertical integration (those renting their chips) will struggle to survive.

The fact remains that even if the financial bubble pops, the technology itself is here to stay. AI models will continue to improve because they are proven to be useful. In my own field of software development, despite being transferred to different departments several times, I still deliver at an impressive pace specifically because of AI acceleration.

I recently did a coding challenge where I implemented Dijkstra’s algorithm in modern C++ and compared it to Gemini and GPT-5.2. Our code was similar, though I believe mine was more maintainable (variable naming-wise). However, it took me 5 minutes to write, whereas these LLMs in high-reasoning modes took just a minute of work to produce similar results.

Agentic AI and Workflow

“Workflow” has been the buzzword of the year, and I expect this to dominate 2026. Currently, AI workflows are not fully mature, but companies have started implementing their own internal systems and adding AI capabilities on top of them.

I expect these workflow systems will mature significantly in 2026, leading to widespread adoption of Autonomous AI Agents. These won’t just chat; they will handle complex, multi-step tasks like “plan this project” or “debug this module” without constant human hand-holding.

We are already seeing this trend with platforms like n8n, Zapier, and LangChain gaining traction in enterprise pilots. These tools are moving beyond simple automation to becoming “orchestrators” that can decide how to execute a task, not just follow a pre-defined script. However, true autonomy still faces reliability and safety hurdles. For 2026, “human-in-the-loop” systems will remain the standard for critical workflows to ensure governance and accuracy.

This evolution will blur the lines between traditional software development and AI-driven automation, creating new opportunities for developers who can orchestrate these agents rather than just writing code from scratch.

Invisible Augmented Reality

Augmented Reality (AR) is currently in its early stages, but I expect it to become seamless and integrated into our daily lives in 2026. The technology is finally becoming “invisible” and context-aware.

We are seeing this with devices using Directional Audio technology (like the Ray-Ban Meta glasses). These glasses have built-in speakers that fire sound directly into your ear canal so only you can hear it—no earphones required.

However, mass adoption faces technical hurdles. Battery life, heat dissipation, and display brightness are significant challenges that engineers are still solving. Until we have batteries that can power a high-resolution AR overly for a full day without overheating, premium models will remain a niche product for early adopters and professionals.

As innovation continues, I expect these invisible AR devices to become more sophisticated. We will move away from bulky headsets toward frames that look like normal eyewear but revolutionize how we interact with digital information.

Unfortunately, I think the premium models will remain expensive for the time being, so mass adoption may not happen as quickly as I’d hope. However, the trajectory is clear: I am looking forward to having less “monitor junk” and LCD screens cluttering my desk because virtual screens can finally take their place—a win for both productivity and the environment.

I know that in VR we can already spawn infinite virtual monitors, but 2026 will be the year this streamlines into lightweight AR that we can wear all day.

AI Scams

As generative AI improves at cloning images and audio, I expect scams to become significantly more sophisticated. Tools like Runway’s Gen-2 and Adobe Firefly have already demonstrated photorealistic capabilities, and open-source models (like Stable Diffusion) can be fine-tuned by bad actors to bypass safety filters.

We’ve already seen issues on Kickstarter, where plenty of projects were created using AI-generated content that later turned out to be fraudulent vaporware.

There have also been reports of delivery scams where drivers use generative AI to create “fake proof of delivery” photos—placing a food bag on a virtual porch using Google Street View data to fool the refund system.

In 2026, we will see new scams that leverage this tech to create increasingly convincing fake content, making it nearly impossible for people to distinguish between real and artificial proof.

We need better digital literacy and verification tools to combat these threats, along with stronger regulations to hold platforms accountable. Detection technologies are advancing, but it remains an arms race. Realistically, next year will be a game of “cat and mouse.” Critical thinking is highly encouraged.

The Blue Collar Boom

Here in the Philippines, AI adoption won’t happen overnight, but we will see a gradual increase in AI-assisted tools for blue-collar workers, particularly in manufacturing, construction, and service industries.

The types of AI systems I expect blue-collar workers will utilize are smart assistants for logistics, AI-assisted information retrieval (repair guides), and decision support tools. Imagine a mechanic using a tablet to instantly pull up a specific engine schematic or an electrician using AR to visualize wiring behind a wall. This isn’t replacement; it’s augmentation.

However, I do not believe blue-collar jobs will be replaced. Even if big tech companies solve their energy problems, the cost of operating AI per request is simply too high to replace human labor in physical tasks.

A perfect example is the Amazon “Just Walk Out” stores. Amazon recently pulled this technology from their large grocery stores because it turned out to be less “AI magic” and more “human reliance.” Reports revealed they had to hire over 1,000 workers in India to manually review the camera footage for accuracy. It was actually more expensive than just hiring cashiers.

This proves that even large corporations are hesitant to adopt full automation when the costs outweigh the benefits. For 2026, the physical worker remains essential, and their value might even go up as digital skills become commoditized.

Conclusion

The road to 2026 promises both massive opportunities and significant bottlenecks. The ongoing RAM crisis underscores the urgent need for sustainable infrastructure. The competition for AI supremacy will be fierce, with companies like Google leveraging vertical integration to gain an edge. But beyond the hardware wars, the way we live and work is shifting.

We are moving toward a world of “Invisible” AR and “Agentic” workflows that handle the heavy lifting for us. However, as AI gets smarter, so do the scams, requiring us to be more vigilant than ever.

In the Philippines and beyond, this shift highlights a surprising truth: as digital skills become automated, the physical world—and the blue-collar workers who build it will only become more valuable.

Introduction

Recently, drama unfolded regarding Larian Studios and AI. It started with a tweet from Kami confirming that Larian is using generative AI in their upcoming game, Divinity.

This tweet sparked an AI “hate mob” on social media, leading to widespread backlash and accusations that Larian is using AI to replace human creativity and craftsmanship. There is also an angle where critics accuse Larian of producing “AI slop,” which I find confusing because their previous work is anything but.

The situation escalated to the point that Swen Vincke, the CEO, had to address the issue publicly. He stated that they use AI mainly to search for references and speed up their workflow. This got mixed reactions. Half accepted his explanation, but the other half stuck hard to saying “NO AI.”

In light of this, I have written this article to share my observations regarding this backlash and to inform readers what “AI slop” really means and when it becomes problematic. It should be noted that I work mainly with AI in B2B markets. This involves using all kinds of AI, not just LLMs but also “traditional” models, such as neural networks that offer deterministic predictability.

Definition of Terms

Before we begin, I would like to clarify the terminology and acronyms I will use throughout this article:

• **AI*– Artificial Intelligence, encompassing all forms of AI, including LLMs and traditional machine learning models.
• **B2B*– Business-to-Business, referring to a business model that caters to other businesses.
• **LLM*– Large Language Model, a type of AI designed for natural language understanding and generation.
• **Neural Network*– A type of machine learning model consisting of interconnected nodes organized in layers.

What is AI Slop?

“AI Slop” is a term coined online to describe low-quality, unoriginal digital content created via generative AI by unskilled individuals. It often implies a lack of creativity, effort, or expertise in the final product. It is typically associated with users who claim AI-generated work as their own.

I observed that this term originated in art circles on X (formerly Twitter), where users would post AI artwork and claim it as their own creation without giving credit or mentioning AI help. Personally, I don’t have a problem with people using AI to generate images, but I do take issue when it’s used to replace genuine creativity and craftsmanship, like in competitions where originality and skill are expected.

Unfortunately, as generative AI becomes more accessible, “slop” has surfaced in other fields such as music, writing, and software development (including the generation of assets like textures and models).

The Backlash Against AI Slop

In art circles, the backlash intensified when people began using AI tools without disclosure. This led to the discovery that these tools were trained on vast amounts of copyrighted data, raising concerns about intellectual property and fair use. This revelation sparked a broader conversation about the ethics of training data and the responsibility of the creators using these tools.

It didn’t help that services like Midjourney trained their models on public sites like Pixiv, where artists shared work without explicit permission. This issue was highlighted by the Midjourney/Artist Database Incident^[1].

Then there was the Amazon incident, which saw a rise in books published using AI-generated content, often featuring nonsensical titles and fake authors (Amazon’s AI-generated book problem^[2]).

In software development, there has been a rise in AI-generated “commits” in open-source projects. These often contain nonsensical code and are pushed by developers who double down on their errors and refuse to acknowledge mistakes, as seen in The Curl AI-generated bug report incident^[3].

Finally, as generative AI improves, companies have begun replacing creative roles with AI. I remember when AI struggled to generate hands or complex structures like a gun, but now companies are pushing it further, such as with the AI-generated intro for Marvel’s Secret Invasion (Marvel’s Secret Invasion AI intro controversy^[4]).

These events have fostered the opinion that AI-assisted work is just low-quality junk, leading to deep skepticism across many creative and technical fields.

Beyond Content Generation

Because of the negative perception of AI-generated content, public opinion of AI in general has soured. It doesn’t help that as generative AI gets better at producing decent outputs, companies are incorporating these technologies more aggressively, like Microsoft integrating AI into Windows. This gives the impression that companies are forcing “AI slop” onto users.

But AI is more than just content generation. It has been used since the early days of computing for automation, optimization, and decision-making. For example, machine learning techniques detect anomalies in system behavior to improve security. This heuristic technology is what powers modern anti-virus software, like Windows Defender.

Even Photoshop featured AI long before the current hype. “Content-Aware Fill” uses AI to figure out which pixels should fill a selected area, showing the utility of AI way before generative models became popular. People also overlook how AI is used in scientific research. From drug discovery to climate modeling, AI speeds up research by identifying patterns at scales impossible for humans. It also powers recommendation systems. That’s how Netflix and Spotify suggest your next favorite show or song.

An Appeal to the Critics

Let’s not throw the baby out with the bathwater. AI has genuine utility beyond content generation, and dismissing it entirely because of poor implementation is shortsighted. Just because a project uses AI doesn’t mean the output is low-quality.

The problem is not the AI itself, but the lack of quality control and responsible use. We need better standards, and I want to help educate people about AI beyond the (completely understandable) “slop” hate.

Will AI take our jobs? It’s a valid concern, but AI is more likely to augment roles than replace them. Many companies that went “all-in” on AI have already begun to regret it, realizing they still need human oversight (MSN replaces human editors with AI then faces backlash^[5]).

Consider automated compliance checking in the B2B sector. LLMs can help draft and review documents to ensure consistency. However, I’ve found that the non-deterministic nature of LLMs is a major hurdle for production use where reliability is crucial. Businesses are hesitant to rely solely on AI for compliance because AI makes mistakes, and someone must be held accountable for those liabilities.

While someone could try to replace me as a developer with AI (and I do use AI as a tool to enhance my work), it is not perfect. It can introduce errors that are tough to trace. An LLM might generate a solid system from a single prompt, but it struggles to maintain that code over time. LLMs have dataset limitations and context window limits. They might not know the latest patterns or get the nuances of a complex, evolving project.

Conclusion

The rise of “AI slop” has created a justified skepticism toward generative tools, but it’s important to distinguish between low-effort shortcuts and meaningful AI implementation. While AI is a powerful tool for processing data and augmenting workflows, its limitations, like a lack of reliability, “hallucinations,” and a finite context window, mean it cannot replace the nuance and accountability of human expertise.

The finite context window has a physical and computational limit that, as far as I know, has no simple circumvention. While some providers attempt to “compact” information or create summaries when the flow hits the limit, these summaries often lose critical nuance (Do We Need More RAM? Is 32GB the New 16GB?^[6]).

Furthermore, VRAM and system RAM remain incredibly expensive. This hardware bottleneck is real; I recently published an article regarding the necessity of 32GB of RAM in 2026, specifically because prices have surged due to AI demand RAM Price Surge: Up to 619% in 2025^[7]. This leads to deeper concerns about whether we are in an “AI Bubble” or, more accurately, a “Black Hole” of investment where massive capital is consumed with uncertain long-term returns (The $60T AI Black Hole Theory^[8]).

Moving forward, the goal should not be to ban AI or mindlessly attack anyone using it. Instead, we must insist on quality control, ethical training data, and the responsible use of these technologies across all industries. That, to me, makes much more sense.

Introduction

As we head toward 2026, many of us are eyeing system upgrades to stay ahead of the curve. I’m in the same boat. I had plans to boost my workstation from 32GB to 64GB of RAM, but with the AI boom, memory prices have skyrocketed as data centers snap up all the supply.

This price jump hits enthusiasts and gamers like me right in the wallet. Even with 32GB, I often find myself wanting 64GB to juggle Podman containers, virtual machines, and daily apps without breaking a sweat.

I’m writing this on my Surface Pro 8, which has 16GB of RAM. It’s still a solid device, but it doesn’t feel like the “high-end” machine it once was. So, here’s the big question for 2026: Is 32GB the new 16GB? Are we at the point where 16GB is just the bare minimum for entry-level setups?

Software Bloat: The “Electron” Tax

It’s no secret that today’s apps are bloated. I remember running Winamp back in 2007, using less than 70MB of RAM when 256MB was standard. Now, Spotify can easily chew through over a gigabyte. What gives?

The main issue is the reliance on browser wrappers like Electron. Developers prioritize speed and cross-platform support over efficiency, so instead of native code for Windows or Linux, they bundle a full web browser with the app.

That’s a hefty memory cost. Run Spotify, Signal, Discord, and Viber at once, and you’re basically running four separate browsers, each with its own overhead. In this scenario, 16GB of RAM gets eaten up fast.

The Shift to Progressive Web Apps (PWAs)

Luckily, there’s a promising trend: Progressive Web Apps (PWAs). As browser tech improves, PWAs let us run apps more efficiently without the burden of multiple standalone browser instances.

I’m hopeful about this change. Major platforms like Facebook, Instagram, and WhatsApp are refining their PWAs. Netflix had a rocky start as Windows users lost offline downloads and yet YouTube proves that feature-packed PWAs can work well by offering such feature to Youtube Premium users.

I’ve switched from the Spotify desktop app to the YouTube Music PWA, and it’s been a smoother ride with much lower memory use.

This efficiency can extend the life of older hardware. I still use a 2015 MacBook Air for everyday tasks, running Chrome OS Flex. Since the OS integrates tightly with PWAs, I can manage messengers (Viber/Signal via Lacros), JetBrains IDEs, and media apps on just 8GB of RAM. It shows that optimized software can reduce the need for constant hardware upgrades.

Gaming: Optimization vs. Brute Force

Gaming, however, is a different story. The demand for RAM is real. Not only do we need more VRAM for graphics, but system RAM requirements are steadily climbing toward 32GB for top performance.

The industry often seems to rely on hardware to cover for software shortcomings. Take Borderlands 4. It struggles even on solid mid-tier setups. When performance issues come up, responses from folks like Gearbox CEO Randy Pitchford often suggest players just accept their hardware limits rather than expect better optimization.

There’s a heated debate about Unreal Engine 5. Is the engine itself unoptimized, or are developers not using it properly? I lean toward developers bearing a lot of the responsibility. Look at Armored Core 6, built on Unreal Engine 4. it runs beautifully even on a GTX 1050Ti, showing that optimization is a choice.

On the other hand, Larian Studios did a fantastic job with Baldur’s Gate 3, optimizing a demanding game for the Steam Deck. Yet, these well-optimized titles are becoming outliers. As AAA games increasingly list 16GB as a minimum, 32GB is starting to feel like the safe bet for serious gaming.

Day-to-Day Usage vs. Power Users

For casual users, not much has shifted. Despite software bloat, 16GB is still adequate for browsing, streaming, and light office tasks. I don’t see an urgent push for everyday folks to jump to 32GB yet.

But for power users managing 20+ tabs alongside creative or development tools, 8GB is outdated, and 16GB is starting to feel restrictive. If memory is a constraint, there are workarounds. take a look at my guide on the Firefox Unlimited Tabs Setup for tips on stretching browser resources.

Conclusion

So, is 32GB the new 16GB?

From where I stand, we’re in a transitional phase. For gamers, developers, and power users, 32GB is the new standard for comfort and future-proofing. Relying on 16GB in 2026 for high-performance tasks feels like a limitation.

Yet, for the average user, the shift isn’t fully here. With component shortages and high RAM prices, the industry might hold off on making 32GB the baseline for budget devices. Still, if you’re building a PC today with a five-year lifespan in mind, 32GB is the logical choice.

Introduction

I first discovered Podman back in 2022. It was listed as one of the supported engines for Distrobox, a tool I was experimenting with at the time. Back then, I was a dedicated Docker user. While I was vaguely intrigued by Podman’s “daemonless” nature, I didn’t feel a pressing need to switch.

I subscribe to the philosophy: “If it works, don’t fix it.” Docker was working for me or so I thought.

My perspective changed when I moved from building simple, small container images to complex, multi-layered ones. I began hitting friction points that turned my workflow into a headache, eventually forcing me to look for a better alternative. That alternative was Podman.

The Problem with the Docker Daemon

The cracks in my Docker workflow appeared during heavy build processes. I encountered a situation where a build failed because my disk reached capacity. In a perfect world, this should just stop the build.

However, with Docker, this triggered an unrecoverable state in the storage driver. Because Docker relies on a central daemon (a background process that manages everything), when that daemon struggles to write layers to a full disk, it can corrupt the state of the engine. I wasn’t just left with a failed build; I was left with a corrupted installation that required me to completely purge my Docker data and rebuild everything from scratch.

This highlighted a critical architectural flaw: The Single Point of Failure.

If the Docker daemon crashes or corrupts, every container it manages goes down with it. It felt fragile.

The Podman Difference: Native Linux Architecture

This led me to seriously investigate Podman. The immediate “lifesaver” feature was its daemonless architecture.

Unlike Docker, which uses a client–server model (the CLI talks to a long-running daemon), Podman works like a traditional Linux command (fork/exec). When you run `podman build`, it is just a process running under your user.

• Stability: If a Podman build crashes due to a full disk, only that specific build process dies. My other running containers are unaffected.
• Safety: There is no central daemon to corrupt. If the build fails, the cleanup is usually immediate and isolated.

In a native Linux environment, this performance is raw and direct. There is no middleman. Podman interacts directly with the kernel’s cgroups and namespaces, making it incredibly efficient for system resources.

Podman on Windows: Escaping the Heavy Desktop

You might be thinking, “This sounds great for Linux, but I use Windows.”

While both Docker and Podman utilize WSL2 (Windows Subsystem for Linux 2) to run containers on Windows, the way they package this experience is vastly different.

Docker Desktop on Windows bundles the WSL2 backend inside a heavy, commercialized application. It runs a resource-intensive GUI and background services that can eat up significant RAM even when idle.

Podman, on the other hand, offers a cleaner approach for Windows developers:

1. Same Workflow as Linux: Podman on Windows runs through WSL2 with the same CLI and behavior you get on a Linux machine. If you develop on Linux servers and use a Windows laptop locally, your commands and scripts stay identical.
2. Lightweight Integration: Because Podman doesn’t force a heavy UI layer (unless you explicitly install Podman Desktop), it often feels lighter on system resources. It leverages the Fedora-based WSL2 backend strictly for the engine, keeping your development environment snappy.
3. You Control When It Runs: There is no always-on “big desktop app” in the background. You start what you need (e.g., a Podman machine) when you need it, and shut it down when you’re done.

The end result: Windows stops feeling like a second-class citizen for containers, and your setup is much closer to a “real Linux dev box” with fewer moving parts.

Simplified Management: The Power of Auto-Update

One of my favorite use cases for containers is hosting Neko^[1], a virtual browser running inside a container. It’s excellent for testing web applications or browsing potentially unsafe sites in an isolated environment.

In the Docker world, updating Neko was a chore:

1. Stop the container.
2. Remove the container.
3. Pull the new image.
4. Re-run the container with the exact same flags as before.

If you manage a fleet of services, this becomes tedious very quickly.

Podman introduces a game-changer called Auto-Update. By integrating with `systemd`, I can simply run:

				
podman auto-update

				
			

Podman checks if a new image is available, pulls it, restarts the container, and even supports automatic rollback if the new container fails to start. It turns a 10-minute maintenance task into a background process I don’t even have to think about.

This approach scales beautifully: from “my one Neko container” up to a host running multiple services that all keep themselves up to date with minimal manual intervention.

Security by Design

Finally, we must talk about security. Docker has historically suffered from vulnerabilities related to its root-privileged daemon.

One illustrative example is CVE-2018-15664^[2]. In affected versions of Docker, the API endpoints behind the `docker cp` command were vulnerable to a symlink race condition. A malicious process inside a container could:

• Prepare a sneaky symlink setup.
• Wait for an administrator to run `docker cp` to copy files in or out.
• Trick the root-running Docker daemon into reading or writing arbitrary paths on the host filesystem.

In other words: the daemon was doing filesystem operations on the host on behalf of a container, with full root privileges. That’s exactly the kind of risk you accept when a central, highly-privileged daemon sits in the middle of everything.

Podman drastically reduces this category of risk through two mechanisms:

1. Daemonless: There is no persistent root process waiting to be exploited in the same way. Each operation is a short-lived process, not a central authority holding open doors.
2. Rootless by Default: Podman is designed to run containers as a non-root user. The default mental model is “my user runs this process,” not “some root daemon runs things for me.”

While Docker now supports “Rootless Mode,” it is often more complex to configure and not how most existing Docker installations are set up. Podman works rootless out of the box, which encourages safer defaults, especially on multi-user systems.

How it Works Under the Hood

Technically, Podman interfaces directly with the Linux kernel’s cgroups and namespaces, adhering strictly to OCI (Open Container Initiative) standards. It uses the same low-level container runtimes (like `runc`) under the hood that Docker does, but without inserting a long-lived daemon into the middle.

The result is a tool that is:

• Secure: Less privileged glue code running all the time.
• Compliant: Built on open standards that play well in the broader container ecosystem.
• Lightweight: Doing only what it needs to, when it needs to, as normal user processes.

When Docker Is Still the Right Tool (Caveats)

I didn’t throw Docker out overnight, and you probably shouldn’t either. There are still situations where Docker makes sense:

• Existing Team Workflows: If your whole team is standardized on Docker, with dozens of scripts, CI pipelines, and docs written around `docker` and Docker Desktop, a migration has a real cost. Podman is mostly compatible but “mostly” still means testing and tweaks.
• Tooling Ecosystem: A lot of third-party tools, tutorials, and examples still assume Docker. Podman’s compatibility (`alias docker=podman`) helps, but some edge cases (especially around Docker Desktop–specific features) may not translate perfectly.
• Mac-Centric Teams: On macOS, Docker Desktop is still the “default” experience many developers know. Podman has solutions (e.g., Podman Machine), but if your org is heavily Mac-based and fully comfortable with Docker, the switching cost might outweigh the benefits right now.
• You Haven’t Felt the Pain Yet: If you’re not hitting daemon corruption issues, not running multi-user hosts, and your threat model is relatively relaxed, Docker might be “good enough” for your current needs.

The point isn’t that Docker is unusable, it’s that, once you’ve seen what a daemonless, rootless-first model feels like, it’s hard to go back.

Conclusion

I didn’t switch to Podman just to be a contrarian. I switched because it treats containers the way they were meant to be treated: as standard Linux processes, not as children of a monolithic server.

We are past the era where we need a daemon to hold our hands. If you value stability, security, and open-source freedom, the question isn’t “Why switch to Podman?” it is “Why are you still tying your containers to a fragile, root-privileged daemon?”.

At the very least, Podman deserves a spot in your toolbox. In my case, it replaced the toolbox entirely.