Andrew Stiefel | essays

AI Isn't Taking Your Job – The Economy Is

2025-05-16T00:00:00-07:00

Have you noticed the latest trend sweeping executive suites? This year, layoff announcements are getting blamed on AI. Or maybe it’s employee performance. No — it’s definitely AI.

But if you look closer, you’ll find a more familiar culprit: the economy.

Companies don’t want to admit revenue is slowing or that fundamentals are crumbling. It’s easier to say they’re “embracing AI” than to admit they’re under pressure to cut costs and protect margins.

Yes, AI is changing how we work — especially for junior roles in software development, sales, content creation, and support. But the wave of layoffs sweeping tech, healthcare, and biotech isn’t being driven by machines replacing humans. It’s being driven by elevated interest rates, post-Covid revenue declines, and investor pressure to do more with less.

AI just provides a convenient cover narrative.

A Correction Disguised as Innovation

Several major tech firms — including cloud providers, social platforms, and enterprise software giants — have collectively laid off more than 600,000 workers since 2022 according to Layoffs.fyi.

These cuts impacted recruiting, HR, operations, marketing, and in some cases engineering. Companies that ramped up quickly during the pandemic suddenly found they were overextended in a post-zero interest rate policy (ZIRP) world. Capital became more expensive. Growth expectations took a cut. And investors needed to see profitability instead.

This is a Covid-era bubble correcting itself. Most of these companies are still larger than they were in 2019. In fact, as research by the Pragmatic Engineer has shown, software engineering in particular has experienced a boom and bust.

The change in the number of listings in 2025, compared to 2020, for each of these areas:

All jobs: +10%

Banking and finance: -7%

Sales: -8%

Marketing: -19%

Software development: -34%

Overall hiring has mostly returned to 2019 levels. There is a hesitation to hire more, but that’s partially due to uncertainty about the level to hire for with AI and ongoing uncertainty about macroeconomic conditions.

From Tech to Biotech: The Narrative Spreads

Like any good pathogen, narratives like to spread. And this one is jumping from Silicon Valley to biotech and healthcare – the other industries most impacted by Covid-19.

A large mRNA vaccine manufacturer recently announced that it was merging the CTO and Head of People into a new Chief People and Digital Technology Officer role. The stated reason? To identify which roles are best suited for AI vs people.

But they are following a now-familiar playbook:

The company reported a GAAP net loss of $1.1 billion loss in Q1 2025
Revenue from Covid-era mRNA vaccines continues to fall
They have reduced operating costs by 19% to compensate

Behind the scenes, it’s facing post-pandemic revenue decline and investor pressure to reduce costs. So they’re turning to layoffs — with AI as the smokescreen.

So what is AI doing to jobs?

There is no question AI is changing work. I use AI in my job nearly every day. And it’s definitely impacting the labor market. A recent study found that:

Junior developers experienced 27-39% productivity gains using AI coding tools.
But senior developers saw smaller productivity gains, typically 8-13%.

AI enhances speed, but doesn’t replace deep expertise yet. In fact, the more you know about a subject or field the less magical it is.

Unsurprisingly, entry-level roles are the first to feel the effects: SEO writers, junior engineers, SDRs, and support teams. But the scale is smaller than the headlines suggest.

Even firms that proudly touted major AI-based role reductions have quietly reversed course — reopening hiring in those same departments just months later.

The Bottom Line

Companies aren’t laying people off because AI is replacing them. At least, not yet. If you’re familiar with AI technology, you know both how far and fast it has come — and all the countless ways it hallucinates or falls flat. It makes a great assistant, but a lousy replacement.

So don’t mistake narrative spin for a technological inevitability.

AI isn’t coming for your job. Because the CFO already did.

Disclaimer: The views expressed in this post are my own and do not reflect the opinions or positions of my employer. Company examples have been anonymized to focus on trends rather than individual firms. All references are drawn from publicly reported data.

Markdown Files, Not Apps

2025-01-17T00:00:00-08:00

Files, not apps. I’m convinced this is the best way to work in the future.

I write down almost everything important in my life: lists, ideas, plans, code, and articles. They form my extended memory. They are a record of what I’ve done, and who I’ve been. That’s why I only use markdown files.

Traditional document and note-taking apps like Word, Evernote, and Notion create data silos and vendor lock-in, limiting your ability to freely move and process information.

I discovered this the hard way when I tried to move between different ecosystems like Evernote and Apple Notes. Sure, there was a way to export my data. But it was a mess. Things broke, attachments went missing, and I spent hours converting, reformatting, and importing.

Markdown is simple

Markdown files solve these challenges through their fundamental simplicity. Any text editor can open and edit them, eliminating dependency on specific applications or platforms. They have all the basic formatting you need for headlines, bullets, lists, and even basic tables.

Markdown is portable

If you use Evernote or Notion, for example, you are helpless without them. If they go out of business you are trapped and have to move your files to another format. You will outlive those companies. I’ve already outlived a few note-taking companies.

I can work with markdown files on any device, forever. I can access them with any text editor I want. This makes them remarkably long-lived. Proprietary formats come and go, but markdown’s open, plain text nature ensures it will likely outlast me. This makes it the ideal format for building a personal knowledge base.

Markdown works offline

There are times when I want to be offline and unreachable, but would still like to jot ideas down in a format that is easily searchable. Working offline is an amazing productivity boost. Why give that up for an online database?

Markdown is ready for AI

This is a use case I wouldn’t have imagined a few years. But it’s the main reason I’m convinced markdown files are the future. Since markdown is just text, AI systems can easily process and analyze your notes.

For example, I can easily grab my “Ideal Customer Profile.md” file and throw it into ChatGPT, Claude, Gemini, or whatever I’m using at the moment to work with it. I can add my messaging framework, and immediately start brainstorming marketing copy.

I can run a local LLM, point it at a folder of my markdown files, and then interact with my notes by asking questions. It’s like a personal assistant built on everything I’ve learned and experienced.

Conclusion

Everything I write is human readable, and yet actionable with apps and tools. I always have my files available locally, and backups are just making a copy of the folder and putting it somewhere safe.

Markdown is simple, portable, independent, and free — exactly how I want to live.

Principles of Platform Independence

2025-01-10T00:00:00-08:00

The internet is filled with stories of people breaking free from Google’s ecosystem (here are a few examples I’ve seen recently). Yet in their rush to escape one tech giant, many users inadvertently chain themselves to another.

I’ve noticed a paradox on social media: people who champion digital independence while relying heavily on platform-specific tools like Apple Notes or Passwords. They overlook how Apple, like Google, builds its business model around service lock-in – from iCloud storage to annual hardware upgrades.

Personally, I prefer to maintain platform independence as much as possible. Over the years I’ve developed a set of criteria — let’s call them principles of platform independence — for choosing what apps and software I use. This way I retain control over my digital life regardless of which company’s software or hardware I use.

1) Cross-Platform Compatibility

Use applications that work across different operating systems (MacOS, Windows, Linux, etc.) to avoid being locked into one platform’s ecosystem. This ensures your workflow remains consistent regardless of the device you’re using.

2) Data Ownership

Choose applications that allow you to easily export your data in formats that can be read by different applications. You should have full control of your data, and it should be accessible in a readable format outside of the application.

3) Data Portability

Beyond just being able to export data, it should be in widely-supported open formats (like CalDAV for calendars, IMAP for email) rather than proprietary ones. This enables interoperability and switching between different clients.

4) Logic Separation

Your data shouldn’t be tightly coupled to specific application features. For example, your notes should be stored in a format that preserves basic formatting even if specific application features aren’t available elsewhere.

5) Local Control

Whenever possible, you should have direct access to your data in a readable format on your local system, not just through a cloud interface. This includes the ability to make local backups. I like to keep a version in the cloud, a version on my local system, a local backup, and a remote back up.

Dead Internet is No Longer a Theory

2025-01-03T00:00:00-08:00

Dead internet theory suggests most of the internet’s content and traffic is artificially generated by bots rather than humans. The conspiracy theory emerged amidst the fallout of the 2016 election, when news broke about websites creating fake news to gain clicks on Facebook and Google and leveraging bot accounts to amplify it.

Ironically, Meta decided to turn the conspiracy theory into reality.

Meta looks into the black mirror

Quoting from 404 Media:

Earlier this week, Meta executive Connor Hayes told the Financial Times that the company is going to roll out AI character profiles on Instagram and Facebook that “exist on our platforms, kind of in the same way that accounts do … they’ll have bios and profile pictures and be able to generate and share content powered by AI on the platform.

It didn’t take long for users on Blue Sky and Reddit to find some of these fake AI profiles on Instagram. The profiles were often offensive caricatures of what a gigantic corporation like Meta might imagine a “proud Black queer momma of 2” might post about. The profiles tended to post AI generated content similar to what other spammer’s on Meta’s Facebook and Instagram platforms have been creating.

Here’s the example I referenced above:

Most of these profiles were already a few years old, and Meta has since announced that they will be taking them down.

Dead internet theory gained traction online because it speaks to genuine concerns people have about authenticity and the increasing use of bots and AI online. I have no idea how anyone at Meta could think launching this exact thing would be a good idea.

I also suspect these profiles will still get rolled out in the future. But next time they won’t have the “AI managed by Meta” labels attached to them.

Navigating the dark forest

So where to from here? We’re entering a new dark forest where it’s increasingly difficult to find the humans among the trees. Maggie Appleton tackled these challenges in one of her blog post’s entitled “The Dark Forest and Generative AI.”

Humans who want to engage in informal, unoptimized, personal interactions have to hide in closed spaces like invite-only Slack channels, Discord groups, email newsletters, small-scale blogs, and digital gardens . Or make themselves illegible and algorithmically incoherent in public venues.

The new challenge isn’t generating content — it’s proving that you’re human.

Fortunately, we humans evolved a whole bag of tricks to tell if something is a predator or not in the forest. Now we have to apply them online. Appleton suggests a few techniques in her post:

Triangulate – verify information from multiple sources (e.g. the SIFT strategy)
Be original – introduce new ideas (AI can only share what’s come before)
Be creative – use language quirks that AI wouldn’t use

In the future I expect we’ll see more institutional verification, similar to what LinkedIn has been rolling out for profiles. In LinkedIn’s case they are using a mix of ID verification (drivers license, passport, etc) to verify people and email verification to confirm companies.

Back to IRL as the norm?

If there’s something to be optimistic about, I suspect it will be the re-emergence of in-person value. Before the pandemic we were already seeing an uptick in “digital essentialism” as people got tired. We were then thrown even more into the online realm with the pandemic, and now we’re seeing more and more misinformation, astroturfing, and other deceptive practices online. I suspect more and more of us will be looking for connection outside the old electronic waterholes of Facebook, Instagram, etc.

I, personally, look forward to a slower, less online world if that’s indeed the future.

The Myth of the Average User

2024-12-31T00:00:00-08:00

This particular bit of history will live forever in my head.

In the late 1940s, the U.S. Air Force faced a deadly crisis: their pilots were crashing at an alarming rate, with 17 crashes in a single day at its peak. The military knew something was terribly wrong, but the cause eluded them.

Enter Lieutenant Gilbert Daniels, who conducted a groundbreaking study of 4,063 pilots. His discovery was shocking: when measuring ten critical body dimensions used for cockpit design, not a single pilot fell within the average range across all measurements. Not one.

This is the “flaw of averages” – the mathematical reality that designing for the average means designing for no one. The average user, like the average pilot, is a statistical ghost.

The Air Force’s solution revolutionized aviation: they abandoned the myth of the average pilot and instead created adjustable equipment. Cockpits with customizable seats, moveable control panels, and adaptable displays became the new standard – a design philosophy that saved countless lives.

This lesson extends far beyond aviation. While I always knew that averages could be misleading (after all, a single outlier can dramatically skew results), I never realized just how fundamentally flawed the concept of “average” could be.

The truth is simple: none of your users are average. Each person who interacts with your product brings their own unique needs, preferences, and ways of working. The key to exceptional design isn’t creating the perfect solution for an imaginary average user – it’s building flexibility into your core experience.

When designing software, you can (and should) have strong opinions about optimal workflows. But you should also embrace edge cases – whether it’s the power user who needs advanced features or the newcomer who needs a simplified interface – to create solutions that work for everyone.

Want to dive deeper into this fascinating story? Check out the original article in the Toronto Star for more details.

Weighing in on AI Hype and Risk

2024-12-10T00:00:00-08:00

Casey Newton wrote a piece last Thursday titled “The phony comforts of AI skepticism” — and wow did people have some thoughts about it:

Gary Marcus responded against the mischaracterization of his position
Edward Ongweso had quite a few more things to say
Casey Newton responded while basically doubling down on his argument
Dave Karpf waded in with his perspective and a recap of the debate

I’m sure there’s more out there but you get the idea. I almost didn’t write something myself, but one part of this conversation still feels like it’s missing to me. So let’s take a brief look at Newton’s argument before diving in.

Newton’s argument in a nutshell

To briefly summarize, the core thrust of Newton’s piece is that the entire range of AI discussion boils down to two groups: critics external to firms and organizations directly working on or studying AI who believe “AI is fake and sucks” and internal critics building AI who understand “AI is real and dangerous” given current and future capabilities. Newton aligns himself with the latter and suggests that the former camp are not only incapable of recognizing genuine innovations made in this field, but risk leaving us blind to threats enabled by those advances.

It’s a widely reductive argument and Marcus and Ongweso do a great job of breaking down the issues with it, and in particular the straw man Newton builds for himself to attack. I’m not going to wade in there, but I recommend reading their posts for context.

I want to talk about how smart journalists like Newton can accidentally (or intentionally) fall for the hype, and in the process fail to correctly frame the conversation for their readers. And I want to be clear up front: I think this was a huge miss.

A false dichotomy

My particular issue with Newton’s article is about the placement of what I’ll call skepticism (“AI is fake and sucks”) and hype (“AI is real and dangerous”) as polar opposites on a spectrum. It’s so wildly off I’m surprised Newton chose to defend it. As many people noted in comments on Mastodon and Bluesky, the skeptics Newton is attacking are more likely to fall into the “AI is real and dangerous and is also fake and sucks” category.

Understanding how current generative AI models, and large language models (LLMs) in particular, fail to deliver on their promise usually means you also understand how dangerous these models are now, and will be in the future. Those shortcomings are especially dangerous when they are released to a public that doesn’t understand what those shortcomings are, and into a media environment that is still grappling with the ramifications of social media.

Sorting out risk from hype

My bigger issue is that Newton seems to fall for the hype from the people building and most likely to benefit from advances in AI. And make no mistake, making unsubstantiated claims about the future dangers of AI is a way of creating hype around the technology. It’s probably the oldest form of hype in computer science – we’ve been talking about imagined AI risks since at least the mid-19th century.

The problem here is that hypothetical risks are treated the same as actual risks. We know that the current GenAI models are accelerants to the issues of propaganda and misinformation that were already widely present on social media and other communication channels before ChatGPT was released. GenAI models are also wrong quite often, despite attempts to improve them, leading to a new form of misinformation when users aren’t appropriately skeptical of the responses they receive. And that doesn’t get into whether LLMs are even a viable path to true general intelligence or not.

We should be talking about emerging risks, but we need a lot more grounding in the conversation from journalists like Newton. And we should differentiate these as hypothetical risks, and they should be treated with more skepticism, especially when they come from the people most likely to benefit from them.

Final thoughts

As Marcus outlines at the start his essay, Newton does get a lot right, including that we should be preparing now for AI to get more dangerous. And we should be documenting the different attitudes that are emerging around AI. I agree with all this.

But we need the media to do a better job of framing the emerging discussion and attitudes towards AI without serving as hype for the individuals and firms building and benefitting from these models.

Just Start

2022-06-28T00:00:00-07:00

How long have you been working on a project with nothing to show?

I am an expert procrastinator. I spent three years building, tearing down, and restarting this blog. The colors were never right, I didn’t like the font, I needed just one more feature to be ready.

Finally, I decided to slim down to just what I need to write. The rest could come later. For now It’s minimal, but it’s all I needed to start.

Today you can publish your writing in less than five minutes. Start with tools like:

The list is endless. So what are you waiting for?

Start writing
Get feedback
Refine your work
Try again

One of the biggest misconceptions about writing or creating anything is that you should start with a strategy: “I’m going to write about X for people interested in Y.”

But here’s the truth: you don’t need a plan to start.

Write about whatever interests you. You’ll eventually find what you enjoy writing about the most and what resonates most with your audience.

You just have to start.

Here are three formats to get started:

Today I learned how to…
Did you read or hear about X? Here’s my reaction…
Someone asked me about X, so here is the answer

Keep a notebook with all the questions people ask you for help with. Write up your answers and publish them.

Eventually you’ll hit your stride — and you’ll probably discover an idea you could never have found while planning your strategy, picking brand colors, and brainstorming topics for your blog.

Start today.

How I Built My Blog with Jekyll and Netlify

2021-12-20T00:00:00-08:00

I’ve been blogging and hosting my website since 2006, but I’ve always been unhappy with the themes available for technologies like Blogger, WordPress, or Squarespace. I usually had a vision for what I wanted to create and would spend hours scouring marketplaces to find something that came close.

At first, I experimented with developing child themes for WordPress. If you’re not familiar with WordPress, I basically overwrote the CSS stylesheets and built my own page and post templates using PHP. I even got good enough that I had clients who would hire me to make changes to their WordPress themes and installations.

Still, all of the tweaks and changes felt cobbled together. They were cobbled together. I was building Frankenstein sites with dozens of overrides, plugins, and hacks that could break at any moment.

Searching for something better

I wanted to escape the bloat of platforms like WordPress, or the restricted design options available with a platform like Squarespace. I wanted to go back to static HTML and CSS and I wanted to code it myself.

After reading online, I came across Jekyll. It’s an open-source static site generator. That means it complies basic text files into static HTML so you don’t have to code every page by hand.

I built a few starter projects with Jekyll and quickly fell in love with it. While there are a variety of open-source static site generators available now, I still enjoy the relative simplicity of Jekyll for getting started. The liquid templating language is easy to understand and your HTML/CSS/JS work cleanly together.

Jekyll may not be the sleekest option — there are good reasons to pick something like Gatsby or Next.js if you work with javascript and react — I still think it is the most approachable option for beginners (like me).

To use a CMS or not?

I built a few starter websites with Jekyll. One of the things I missed at first was the integration with my writing tool of choice (Ulysses). I also wanted some of the benefits of a CMS, like scheduling posts. Ulysses integrates with WordPress and Ghost so you can publish from mobile and desktop. I decided to give Ghost a try.

As a CMS, I did love Ghost. It’s simple and focused on the writing experience. They have even embraced the Jamstack (Javascript, APIs, and markup) movement driving static websites and supporting headless designs. I also liked the idea of keeping my content separate from the frontend framework.

I worked with Ghost for a while but kept getting frustrated with the server upkeep that it invariable entailed. At first, I paid for a monthly subscription, but the functionality at the basic level is limited unless you spend $30/mo. At that level, you can either create your own theme to host with them or use their API to support a frontend framework built with Jekyll, Gatsby, or another tool. The cost is about the same as a good WordPress hosting provider, but that was more than I was willing to spend for the convenience of scheduling posts within a CMS.

I should note that Ghost also offers an option to self-host (starting at $5/mo through Digital Ocean) but I didn’t want to maintain the server myself. I already have enough reasons to not write without throwing in server updates…which were one of the reasons I wanted to leave WordPress.

Ultimately, I decided to forgo the CMS and focus on Jekyll. I’ll write more about this in the future, but Jekyll fits into my growing philosophy around working with markdown for personal knowledge management and publishing.

Publishing with Netlify

My first Jekyll website was published using Github pages. While I loved the option to publish a free project website directly from my repo, it was definitely more difficult to manage and came with a host of limitations.

Ultimately, I went with Netlify. They are a fantastic organization that makes it easy to build, deploy, and scale web projects. Using Jekyll, I can generate a website from my repo on Github and Netlify deploys it to edge servers across its network. Since I’m only serving static HTML files, this makes my websites incredibly fast for readers. And there is no server (on my end) to worry about securing, protecting, etc. Netlify handles the build and deploys the static files acrss the edge.

Getting set up

I won’t go into too much detail on the development process here, but I will link to the resources that I found the most helpful for getting started:

Poole: Clean and concise foundational setup for Jekyll
How to deploy Jekyll with Netlify
How to deploy Jekyll with GitHub pages
Formspree: Open-source form solution for static websites
Liquid template language reference from Shopify
Getting started with Netlify

As one of the original static website generators, Jekyll benefits from a large community of open-source contributors and users who have documented their techniques and solutions. If you get stuck, there is most likely a tutorial with the answers you need!

What’s next

Overall I’m pretty happy with the results of the project. I learned a lot along the way and ended up with a website that doesn’t look too bad and is unique to me. While I don’t need to make any changes to start writing, there are a few areas I plan to explore in the coming months:

Implement Tailwind CSS to replace Bulma I used the Bulma CSS Framework to scaffold the development of this project. Previously, I used Bootstrap for a few starter projects. Bulma is great, but I’ve been wanting something easier to customize. Enter Tailwind CSS, which is a just-in-time, atomic framework for creating CSS. I can build and design the website, all within HTML. Then, I can filter out any unused classes, resulting in a very small file.

Update 2024-12-01: I completed this work back in May 2022. I am currently using the excellent jekyll-tailwindcss plugin to implement Tailwind CSS for my site. I wraps the Tailwind CSS CLI and lets you specify which version of Tailwind CSS you want to use.

Buildout functionality for my digital garden I expect to write about this more in the coming months, but I’m planning to treat this blog more like a digital garden. There are some core functionalities I want to add, like sides notes, bi-directional links, and search, to make the reading experience more rich and interconnected.

Update 2024-12-01: I’m still working on this. Overall this feels a bit more like a distraction from writing than a priority. I have played around with introducing categories for evergreen, seedling, and budding notes to indicate how complete an idea/post is.

Rebuild the site with Gatsby, Next, or another framework I really like Jekyll, so this one isn’t urgent. But I’m starting to see how javascript frameworks like Gatsby, Next, or Vue could enable some new capabilities. One function that stands out to me is the ability to import markdown (my blog posts) from a separate Github repository. This way my thinking and writing could truly exist separately (and under version control!) from the frontend visual design.

Update 2024-12-01: I no longer plan to do this for a variety of reasons. I may write more about why in the future, but the short version for now is that I hate dependencies and heavy frameworks. I like how simple Jekyll is to run, and that the final result is really just a bunch of html files, one stylesheet, and one bit of non-critical, vanilla Javascript.

I hope you have a similar positive experience, and please reach out if you have any questions about building your website with Jekyll and Netlify.