John’s Blog

I love working on the command line. It’s 2026 and we have some of the best tools in the world for building amazing graphical user interfaces, but here I am at the command prompt like it’s 1983. The command line isn’t old, it’s eternal and has always been the most direct way to control a computer.

The command line is powerful, but it’s cumbersome. Many of the best CLI tools are open source or hobby projects where documentation is wildly inconsistent: nonexistent at worst, overwhelming and deeply technical at best. For something I use every day like git, that’s fine. But many CLI tools are purpose-built utilities that do a few things extremely well, and I don’t use them often enough to internalize their syntax.

Enter the LLM. My command line awakening has been using Claude Code as a front-end for my CLI tools. It turns out LLMs are uniquely well-suited for this. The major open source tools are well-documented in plain text — blog posts, Stack Overflow threads, GitHub pull requests, readme pages — all of it embedded in the model’s weights. And CLI syntax is predictable and structured in a way that natural language isn’t, with deterministic failure modes that give the LLM real-time feedback on what worked and what didn’t.

Managing Google Workspace is a perfect example. The admin interface is cumbersome and riddled with gotchas, and I’ve always avoided it for fear of breaking something. But running a growing company means you can’t ignore it. I found a tool called GAM, a command line utility that handles everything a Workspace admin needs (users, domains, aliases, groups) straight from the terminal. It’s remarkably well-documented. Though honestly, you don’t need to read any of it. That’s the whole point.

I’ve never learned a single GAM command, but I’m using it weekly to reset passwords, manage aliases, adjust settings, and handle the administrative work that used to eat up my afternoons.

Here’s a simple example of something I’d sent to Claude:

Hey Claude- could you reset the password for [email protected]?

A few seconds later:

Done. Password reset for someone@airmail:

- Temporary password: <a temp password here>
- Force change on next login: enabled — they'll be prompted to set a new password the first time they sign in.

That alone is worth it! No more navigating 12 clicks deep through the sluggish Google Admin website. But then Claude volunteered something I hadn’t thought to ask:

A couple of follow-ups worth considering:

- 2FA status — if they got locked out because they lost their 2FA device, the password reset alone won't help. Want me to check their 2FA enrollment? ./bin/gam info user [email protected] will show it.
- App passwords / active sessions — a password reset doesn't automatically sign them out of existing sessions or revoke app-specific passwords. If this reset is for a security reason (suspected compromise), let me know and I can also run gam user [email protected] signout to terminate active sessions.

That’s the part that hooks me. It’s not just executing commands … it’s bringing operational awareness I wouldn’t have had otherwise.

I’m also building my own CLI tools now (using Go with Cobra, Viper, and Charm) to manage my apps, services, and infrastructure. The interesting shift is that I’m designing them knowing Claude Code is the front-end. Clean help text, consistent output, predictable failure modes. Not optimized for a human memorizing flags, but for an LLM that will read the docs once and orchestrate from there. It’s a different way to think about tool design.

It’s a good time to be back at the terminal. ✌️

I first started using Heroku somewhere around 2010. At the time I had managed almost all of my projects on either corporate infrastructure or physical hardware we had purchased at a hosting provider. Writing those words in 2026 sounds completely archaic and byzantine.

Yes, we purchased physical server hardware and had someone manually set it up so we could connect to it remotely. The “cloud” was barely a thing at this time, and certainly not something that we took seriously for “important work projects.”

Right before the Heroku days I had my first project that went viral, as we’d call it now, and we had a major hosting crisis. I distinctly remember my setup for this launch: I had 2 beefy dedicated web servers and a dedicated database server. We thought it would have some traffic so we planned early.

As soon as I switched over the DNS server to push this site live we were instantly flooded with traffic. I found out later that the domain for the project, a little promotional app for a chicken sandwich company, was included on in-store signage that was posted too early.

Here’s the fun part that seems quaint now: we had to order more physical hardware to account for the traffic. I tripled the capacity, which was largely a guess at how much we’d need, and placed a rush order for new hardware and setup. The “rush order” gave me that new hardware within a few business days and I received the login information to get connected. Oh yes, I couldn’t actually even use this new capacity until I configured the servers, by hand.

I was smart enough at the time to have scripts that would mostly automated the configuration of the servers, but it still took hours of time. I had to install a web server (Apache, I assume), configure our middleware layer (mod_rails, which was later renamed to Phusion Passenger), set up Ruby, configure the firewall, and I’m sure a number of other things on each server before it was ready to go. Within a few stressful hours I deployed the new capacity and switched our load balancer (also on dedicated hardware) over to account for the new setup. We continued to burn hot for a few weeks until the promotion was over but thankfully didn’t need to order any more hardware after that.

This wasn’t a unique event, but it was indicative of how I approached servers and hosting at the time. Imagine my joy when I was first introduced to Heroku, and this was all you needed to do to deploy your app:

git push heroku master

That was it. Literally one command in the terminal to replace what my silly scripts would do for me in hours. By default, Heroku would create a small slice of a server (they called it a Dyno), install everything needed, and spin up a web server to handle running your code within a few minutes.

If you were gaining traffic or needed to scale your service it was as simple as switching the dyno count from 1 to 2, and so on. Incredible. What took me days only months prior would now take minutes.

I switched everything I was working on to Heroku and didn’t look back for almost the next decade straight. It certainly had its problems over the years, but in general I loved Heroku.

Fast forward to this week, and the following message was posted to the Heroku blog:

Today, Heroku is transitioning to a sustaining engineering model focused on stability, security, reliability, and support. Heroku remains an actively supported, production-ready platform, with an emphasis on maintaining quality and operational excellence rather than introducing new features. We know changes like this can raise questions, and we want to be clear about what this means for customers.

There is no change for customers using Heroku today. Customers who pay via credit card in the Heroku dashboard—both existing and new—can continue to use Heroku with no changes to pricing, billing, service, or day-to-day usage. Core platform functionality, including applications, pipelines, teams, and add-ons, is unaffected, and customers can continue to rely on Heroku for their production, business-critical workloads.

Enterprise Account contracts will no longer be offered to new customers. Existing Enterprise subscriptions and support contracts will continue to be fully honored and may renew as usual.

We’re focusing our product and engineering investments on areas where we can deliver the greatest long-term customer value, including helping organizations build and deploy enterprise-grade AI in a secure and trusted way.

First, this is an incredible exercise in big company PR saying something without actually saying anything at all so I’m leaving a lot of this to speculation. But it seems very clear to me that this is Heroku being put into maintenance mode. The infrastructure will continue to be supported and kept functional and secure, but I wouldn’t expect to see much more than incremental updates from here on out. I’d love to be wrong, but the writing has been on the wall for some time.

Here’s the real indicator to me: “Enterprise Account contracts will no longer be offered to new customers”. Enterprise contracts are crucial to support a growing organization and likely represent the most important and highest revenue customers. (I know for a fact that when you reach a certain monthly spend amount on Heroku, you would trigger a sales person to reach out and attempt to convert you to an enterprise account.) Without new enterprise customers and the sales staff that support them, the platform is dying. Sure, the business will likely maintain its current course for years to come but without growth it’s a lost cause. I could be wrong, I hope I’m wrong, but it sure doesn’t seem like it.

Maybe it’s time. Like I said, the writing has been on the wall for years. Heroku was never the cheapest hosting option, but over the past 6 years it became almost laughably expensive. The Heroku model has been proven and it is a viable business. Just not for Heroku itself it seems.

There are a number of excellent competitors that have been founded recently and taken what Heroku pioneered to new heights. Fly.io, Vercel, Engine Yard, Render, and many more.

I migrated Air Mail off Heroku last year to Render.com and couldn’t be happier. We cut our bill in half, and increased capacity and throughput at the same time by almost 3x. The developer experience is close to what Heroku offered (alas, I still miss pipelines) and getting better by the month. It’s wonderful to be on a platform that is evolving and growing. The Render support team was incredible as well: they helped me move my many-hundred-gigabyte Postgres database from Heroku to Render with only about 30 minutes of production downtime.

I’ve seen chatter that the Salesforce acquisition of Heroku was to blame for its downfall. Maybe, but I’m not so sure. Salesforce has owned Heroku since 2011 and by all accounts it seems to have fueled Heroku’s growth for many years. At a certain point the decision was made that Heroku wasn’t the future of developer-ops and hosting for Salesforce and their interests went elsewhere. I don’t blame the approach, but that doesn’t make it less of a shame to lose a once great product.

So let’s pour one out for Heroku: You had a great run. You made a lot of engineers out here happy for many years. You introduced us to the concept that cloud hosting isn’t bad. You helped make cloud-hosting approachable and predictable. You held on to that profit margin for servers as long as you could. You led the charge to make better web apps. You changed the game.

Cheers 🥃

Like most weeks over the past year, my mind is deep into the world of AI. It’s hard to escape, and I’m finding new ways to use these tools by the day.

This week saw the announcement of Xcode 26.3 with its first offering of agentic development tools and it looks incredibly cool. I’m not a big Xcode user these days, as I’m not currently working on an iOS app, but I’m never too far from it. Apple has been, let’s say, slow when it comes to adopting new AI features and paradigms so this is a big one. For those that live in Xcode, this must be a huge announcement and a welcome change.

My usage of AI for development and utilities has primarily centered around Claude Code. I’m a command line guy. I’ve always felt comfortable working within the shell and I love having all of my tools in one place. A few coworkers have tried out Codex and the more GUI-friendly interface for Claude Code, Cowork, but I’m finding it harder and harder to leave Claude Code behind. Especially for workflows that aren’t even related to “code”.

Most of my daily non-development work lives in Obsidian. At its core, Obsidian on my Mac is just a folder full of Markdown files. Daily notes, lists of todos, meeting notes, project documentation, and all of my general thoughts and ideas around my work lives in my Obsidian vault. It’s all in plain Markdown, with lots of tags and properties (metadata) to keep things organized. It sounds complicated, but it’s incredibly simple and basic.

There’s no official support for an LLM or AI tooling within Obsidian. Again, it’s just a folder full of markdown files. This is where Claude Code comes in for me. I use Code on a daily basis to interact with and automate my Obsidian vault and do the heavy lifting for me and keep everything organized. Not only is it saving me a ton of time, it’s incredibly satisfying to kick off the routines and then focus on something else for a while.

The Daily Routine

Each morning I have a /daily command that sets up today’s daily note. The note is set up, linked back to my weekly note for that week, and my day’s agenda is added via a connection to my calendar app. I’m in way too many meetings for one human on a daily basis, so it’s helpful to have my AI assistant prepare what’s needed for the day.

Each meeting on my calendar also gets its own note file. Those individual notes are all set up and linked back to the daily agenda all from Claude without me having to open anything, or even look at my calendar.

In the case of recurring meetings, Claude automatically links up the previous meeting of the same kind, and then in that previous meeting links to today’s new occurrence. So in each meeting note file I have a convenient “Next” or “Previous” meeting link to go back and forth between each week.

For a 1:1 meeting, Claude summarizes the last few 1:1 calls I’ve had with that particular person and gives me a quick list of what we’ve been discussing recently. I also have Claude connected to a bunch of other tools so I can see progress made and what each team member has been up to, but that’s a topic for another post.

After each meeting, or at the end of the day, I’ll compile my handwritten or typed notes and save them in the note file for next time. If there’s a meeting where we have a transcript, that’s added and automatically included as well. (I’m using Hazel on the Mac to drop transcript files in a folder and automatically send them to OpenAI for summarization.)

This system is perfect for me. No information is lost, all progress is captured, and everything is indexed and linked together. Every bit of context throughout my week is captured and organized, so I don’t have to keep all of that information in my head. And since Claude is linking everything together automatically, I have full context on each project, meeting, person, and task I’m working on. Lovely.

All of this work is done within a Claude Code window in my terminal. I run a few commands and Claude takes care of the rest. I’m just scratching the surface here and there’s so much more, but this is just a little bit on how I’m using Claude without even writing code to get things done.

I’ve never been the biggest college football fan. I didn’t go to a huge state school with a big team. I didn’t grow up in a college football family. The NFL has always just spoken more to me starting as a kid as a Bills fan before the Ravens showed up in town in 1996.

For me college football has always been a bit sloppy. These are kids out here, after all, playing a very difficult and physically intense game. The quality of play compared to the NFL has always been harder for me to watch. Missed throws, dropped passes and poor defense isn’t an entertaining sport in a time when there are endless entertainment options available.

The majority of the ‘big’ games have traditionally been lopsided blow-outs with big school power players always having the advantage. I’m not interested in watching a giant school with a dozen future NFL starters beat up on a small school by 50 points week after week. It’s just not that interesting.

This week the Indiana Hoosiers won the College Football National Championship. Yes, the traditionally basketball-focused school outlasted and outclassed some of college football’s most storied and powerful programs. Prior to this week Indiana hadn’t won its conference since 1967, and had never previously won a playoff game. And this year they went undefeated to a national title.

The sport is feeling different. It’s more like the NFL, and that’s a great thing. Players are being directly paid to play the sport through NIL money which means there’s no rush for the best players to jump to the NFL to cash in on their abilities. There were multiple players 23 and 24 years old playing on the Miami team that lost in the championship: unheard of prior to the NIL for a good player. Good football is a mental sport and keeping players in the system longer to gain more experience is making the players, and the sport, better.

The ‘transfer portal’, which is a super cool name for players being able to transfer to different schools if they desire, has been much maligned as ruining the sport as well. There’s no loyalty to schools, the critics say. They said the same thing when the NFL embraced free agency decades ago. They were wrong then, and the critics are wrong now.

The sport has evolved, and it’s okay. It’s better. The competition level is higher. The players are staying around longer and building their skills before making the jump to the NFL which is making both college football and the NFL better in the long run.

The money being funneled into college programs is a good thing for the sport. I love that Marc Cuban can donate his money and time to make his alma mater better. I hope there’s much more of this in the years to come. It’s not just Indiana, but they’re proving the model and they are writing the book on how to succeed in modern college football.

I’ll still be mostly watching football on Sundays. But change in college football is good. It hasn’t been great for a long time, despite some fun regional rivalries and historically interesting matchups. Indiana’s year was so fun to watch and it gives me a lot of hope for the future. The TV ratings (this was the most viewed title game since 2014) and the online discourse shows that the general public agrees with me. College football has evolved, and we’re here for it.

Have a great weekend. ❄️

For the next few months I’m going to be very consumed merging two companies together into one tech stack. One of the key reasons we were acquired last year was to unify a technology stack that has helped our business grow and to bring the tools we created to the larger Puck enterprise.

Both legacy organizations are running completely independent technology at the moment, but my goal is to bring it all together. Puck’s business is primarily run on Wordpress, which is a lovely platform for basic CMS needs but not great for running a complex editorial business out of the box. Quite frankly, it’s capped the growth trajectory of the company and I’ll be fixing that very soon.

But before we even think about content and publishing, I need to unify the billing and subscription side of the business. Both companies are running separate billing systems, tax systems, and reporting back-ends. And, both are single tenant by nature: they were purposely creating to run a single business. So we’ve kicked off the work of creating a multi-tenant architecture that will house legacy Air Mail, legacy Puck, and any future acquisitions all in one place.

This effort has been a ton of fun, and the team seems to be enjoying the ability to develop on a new system with some greenfield requirements and shed some of our legacy baggage. I know I feel that way.

We’re just getting started but the effort is already moving along at a steady pace. I love starting fresh and designing the architecture for the next 5–10 years of this company.

Elsewhere

In sporting news, the NFL playoffs have been such a joy this year. It’s refreshing to not have a single rooting interest so I’m able to actually enjoy the games for face value and root for the best scenarios, instead of always scheming for my team’s success.

I’m delighted to see John Harbaugh agree to a 5-year deal with the Giants to become its next head coach. I’m also delighted he’s not in the AFC. I’ll be rooting for him for sure.

Anthropic announced its Claude Cowork features which bring the power of Claude Code to more non-technical users that don’t want (or know how) to live in a terminal window. Claude Code is my absolute favorite tool right now and I’m using it constantly, for coding and non-coding projects alike. Excited to give Cowork a spin, even though I know I’ll probably keep using Code for now.

Apple announced the Apple Creator Studio which brings together Final Cut Pro, Logic Pro, Pixelmator Pro, and a bunch of other tools into one purchase and one subscription. $12.99 per month or $129 per year seems like a great deal for these apps. Not that I need another Apple subscription, but this looks very attractive.

OpenAI explained its approach to ads within ChatGPT which has felt like an inevitability given how much money they are spending. The framing is thoughtful and coherent and seems to be well considered. I’m still more than happy to pay for an ad-free subscription because of the immense value I’m getting from ChatGPT, but I completely understand the desire to open up the market to even more users that are more price sensitive and don’t mind seeing ads in exchange for using the service.

Happy Sunday. Here’s to another great week ahead. 🏈

A week into the year and I figured it’s as good a time as any to pick back up with some light blogging. The last half of 2025 was a complete blur with so many emotional ups and downs it’s not worth recounting them all. Overall I’m incredibly thankful and grateful for the chapter of life that I’m in. In short, I’m stoked for the year ahead.

This morning I updated my LinkedIn for the first time in about 7 years. I don’t like it over there (and I hope to go another 7 years without returning) but I made the jump to “formally” add Puck to my profile.

Late last year, Puck completed its acquisition of our little media startup called Air Mail. I’ve joined Puck as its CTO and couldn’t be more excited to have started this new chapter. The great news is that Air Mail is still alive and thriving within Puck and I still get to work on it every day.

🎧 Speaking of Puck, our incredible CEO Sarah Personette joined The Rebooting podcast with Brian Morrissey and shared a ton of great info about our approach and where we’re heading this year. (A nice shoutout to yours truly towards the end was also very cool of her.) Listen on Apple Podcasts or Overcast.

As we begin the process of merging two completely different company technology stacks, I’m knee deep in modeling and understanding both businesses so we can unify everything this year. We’ll be using Air Mail’s technology as the basis but making some substantial upgrades and improvements as we go. More on that throughout the year.

One week down. Here we go, 2026…

Yesterday the news broke that the Ravens had fired coach John Harbaugh after 18 seasons with the team. For context: 18 seasons is an eternity in the NFL and it's a credit to him and the franchise that they have been able to have such stability and success over this era. Harbaugh is for sure a future hall of famer and he should be recognized for how great he has been, and will continue to be for wherever he goes next. I'm sad about this news, but that's the business of the NFL. It's time to move on, and we'll see what happens next.

Thanks for the run, coach. 🙏

🎉 Happy New Year! I'm excited for a clean slate and to get back to work building. 2025 was a long one, but we made it through. Onward!

Is the age of AI-powered web browsers upon us? Another entry announced earlier this month is Perplexity’s Comet browser. Like Dia, it looks really nice and I’ll look forward to giving it a spin. (I’m on the waitlist, alas.)

The demo video from the announcement on X has a cool walk through.

Dan Goodin, writing for Ars Technica:

Authorities and researchers are sounding the alarm over the active mass exploitation of a high-severity vulnerability in Microsoft SharePoint Server that’s allowing attackers to make off with sensitive company data, including authentication tokens used to access systems inside networks. Researchers said anyone running an on-premises instance of SharePoint should assume their networks are breached.

The vulnerability, tracked as CVE-2025–53770, carries a severity rating of 9.8 out of a possible 10. It gives unauthenticated remote access to SharePoint Servers exposed to the Internet.

I worked on a few SharePoint projects in the early 2000s, and even back then it felt old and outdated. I was shocked that anyone would choose to use it for any purpose. And yet, here we are in 2025, and I have the same reaction anytime I see the word SharePoint crop up.

Last night’s Major League Baseball All Star Game was fantastic. The MLB has done a remarkable job taking this sport, which is rooted in century-old tradition, and modernizing it while making it entertaining. The other major leagues should be taking note of how baseball has improved and follow the lead.

Clayton Kershaw was mic’d up while pitching during the second inning and it was so much fun. “I’m gonna try and throw some cheese real quick hold on” he said as he thew a pitch. He even solicited the help of the commentators on which pitches to throw. Pure genius mic’ing Kershaw up during a game.

The league introduced the usage of ABS (Automated Balls and Strikes) that allowed either a batter, pitcher, or catcher to call for a review of an umpire’s ball or strike call. The players loved it, given their smiles while requesting its usage, and the system works: it correct a few incorrect calls.

The game technically would have ended in a tie, but the league had a great solution for avoiding that completely un-American result: a swing-off home run contest.

Watch the faces of the players competing in, and watching, the swing-off and tell me they all weren’t having a blast.

A fantastic night of entertainment. This is how you do it.

Kite is a fresh take on news consumption by the team at Kagi:

Kite reads public RSS feeds of thousands of (community-curated) world-wide news sources and distills them into one perfect daily briefing. You get every critical perspective and timeline in just 5 minutes. That’s it. No endless scrolling. No attention hijacking. Because we deserve better.

The public files for Kite are all open source on Github.

Kurt Wagner, for Bloomberg:

Meta Platforms Inc. Chief Executive Officer Mark Zuckerberg announced a major restructuring of the company’s artificial intelligence group, including a commitment to developing AI “superintelligence,” or systems that can complete tasks as well as or even better than humans.

Zuckerberg wrote Monday to employees that Meta’s AI efforts will fall under a new group called Meta Superintelligence Labs, or MSL, which will be led by Alexandr Wang, the former CEO of data-labeling startup Scale AI, according to an internal memo reviewed by Bloomberg. Wang, whom Zuckerberg called the “most impressive founder of his generation,” will serve as chief AI officer. […]

Meta will spend “hundreds of billions” on AI projects and research in the years to come, Zuckerberg has said, though the Facebook founder also expects that many firms will likely overspend on AI in an effort to avoid missing the wave.

CNBC has a copy of the memo from Zuckerberg, which begins with this:

As the pace of AI progress accelerates, developing superintelligence is coming into sight. I believe this will be the beginning of a new era for humanity, and I am fully committed to doing what it takes for Meta to lead the way. Today I want to share some details about how we’re organizing our AI efforts to build towards our vision: personal superintelligence for everyone.

Emphasis above is mine. What a bold statement.

Olivia Sophie Rafferty, for ai fray:

The Northern District of California has granted a summary judgment for Anthropic that the training use of the copyrighted books and the print-to-digital format change were both “fair use” […]

However, the court also found that the pirated library copies that Anthropic collected could not be deemed as training copies, and therefore, the use of this material was not “fair”. […]

This is a mixed ruling on fair use – a loss for both copyright holders and Anthropic, but potentially a big win for AI platforms in general. And, if upheld, the order would mean that AI firms using copyrighted material to train their LLMs may be allowed in the future. The only exception to this would be if the material has been pirated.

The mixed ruling in this case is potentially very interesting for the larger AI and LLM-based industry. The court here is establishing some precedent that training an LLM model is considered ‘transformative’ because it uses existing copyrighted works to create new outputs. This ruling sure seems to validate the approach of training LLMs on large datasets of copyrighted materials as acceptable, in the eyes of the court at least.

On the other hand, ‘pirating’ a bunch of books and content is obviously not fair use, and Anthropic will be on the hook for those damages in a future trial.

Hayden Field, for The Verge:

OpenAI has scrubbed mentions of io, the hardware startup co-founded by famous Apple designer Jony Ive, from its website and social media channels. The sudden change closely follows their recent announcement of OpenAI’s nearly $6.5 billion acquisition and plans to create dedicated AI hardware.

OpenAI tells The Verge the deal is still happening, but it scrubbed mentions due to a trademark lawsuit from Iyo, the hearing device startup spun out of Google’s moonshot factory.

Turns out that one of the hardest things in computer science still applies to our AI-driven world: naming things.

And really: “io” is such a common phrase in computing, it shouldn’t be able to be trademarked at all.