Follow ZDNET: Add us as a preferred source on Google.
ZDNET’s key takeaways
- ChatGPT Pro delivers nonstop coding assistance.
- Context switching, once a bottleneck, disappears.
- Marketing now takes longer than development.
Four years of product development in four days, for $200. That’s a pretty hefty claim. But, to my astonishment, it’s true. It’s also a matter of perspective. The math that works for me might not work for you.
Obviously, this is an AI story. The $200 is the ChatGPT Pro plan, to which I upgraded to get more cycles out of the OpenAI Codex programming-tuned AI model. The results can only be described as extreme.
Also: 4 ways I save money on my favorite AI tool subscriptions – and you can too
I need to offer some context to help you understand how I can make this claim and how it fits into my world of work.
My backstory
I am not a full-time programmer. I am a formally trained computer scientist. I taught programming at the college level for years, but I don’t make the majority of my living writing code. Over the years, I’ve seen my peers from college lose their edge as they left programming behind to move onto managerial responsibilities.
My career track has been similar, but I didn’t want to fall behind on my technical skills. So, I’ve always made sure to have a variety of technical projects going and regular continuing education training to keep my chops up.
Almost 10 years ago, I was looking for a new technical project. I came upon a donations system based on WordPress. The original developer could no longer keep it maintained. I acquired it, and converted it to a freemium model: a core product available as free and open source, and a variety of value-add add-ons that brought in some additional money.
I also acquired a personal security product, also with an existing installed base. Its developer was also no longer able to maintain it. This, too, I converted to a freemium model.
Also: I found 5 AI content detectors that can correctly identify AI text 100% of the time
All told, I adopted 10 WordPress plugins, but I only converted two of them to freemium, with paid add-ons. When I first adopted these, I assumed all WordPress site owners were technically skilled, and all I would have to do was provide basic security updates. I was wrong. Most of the users had almost no technical experience, which meant I inherited a giant support load.
Keep in mind that I had a full-time job, and this plugin endeavor was very much a side gig. At best, the freemium products covered the costs of supporting everything. Monthly licensing fees, for everything from the mailing list manager to hosting to the help desk software, have been coming out of my pocket every month since I adopted the plugins.
Because this was a side gig, I really didn’t have all that much time for ongoing product development. For me, coding always requires a mindset change. It also always requires coming up to speed with the development environment and the codebase, plus incorporating any inevitable mandatory updates that change how everything works.
Also: AI models know when they’re being tested – and change their behavior, research shows
I couldn’t just pop in for 10 minutes here or there when I had a gap in my schedule. I usually needed at least 2-4 days of unbooked time to make any progress. Over the course of a year, I’d get maybe three or four such open blocks of time.
Net-net-net, in the 10 years I owned my collection of open-source products, I produced 10 add-ons, ranging in price from $20-30. Most were fairly small feature enhancements. Only one, for my security product, was substantial. Remember that 10 add-ons in 10 years statistic. We’ll come back to it.
I maintained all of the core open source for about eight years, but having so many plugins meant I had difficulty focusing on any particular one, especially given the time constraints of this being a very side gig. Last year, I decided it was time to focus. I sold off all but one of the open source products, the security product.
Getting started with Codex
Last month, I tried connecting ChatGPT Codex to my security product’s code. At that time (in AI terms, it was years ago), Codex only worked in the cloud, on GitHub repositories. I eventually found it to be somewhat functional, but tedious.
Also: Why I’m switching to VS Code. Hint: It’s all about AI tool integration
Earlier this month, OpenAI added the ability for Codex to run in VS Code, integrated into the development environment. This is where I like to work.
Codex in VS Code is sweet. It works quite well. Yes, it does bonehead moves, as do all AIs. But because it is so nicely integrated into VS Code, undoing those bonehead moves and trying another approach turns out to be swift and easy.
Codex in VS Code is available in the $20/month ChatGPT Plus plan. For $20/month, Codex does quite a lot, but I found that it definitely does throttle productivity. At one point, it cut me off for 30 minutes. After about 5 hours, it cut me off for 90 minutes. And then, after another 5-hour run, it cut me off for almost 6 days.
Also: I did 24 days of coding in 12 hours with a $20 AI tool – but there’s one big pitfall
That was game over.
During the time ChatGPT Plus allowed me to use Codex, I got a lot done. I calculated (and you can read the article for details) that I got about 24 days of work done in 12 hours. I used it to make a bunch of enhancements to the core open-source security product, which I plan to release fairly soon.
Taking the $200 plunge
Last Wednesday morning, I was cruising into one of those very rare and precious open time blocks. I had completed all my articles for the week. I didn’t have any must-do work scheduled for the rest of the week. This week I do. This week is packed.
Looking at my calendar, I realized that I could allocate Wednesday afternoon through Saturday evening to project time. I had just finished (finished, meaning I was cut off and locked out until Friday) using Codex to do some product enhancements.
Also: Gemini just aced the world’s most elite coding competition – what it means for AGI
Working with Codex on those product enhancements had been amazing. I really enjoyed the rush of getting work done that quickly, even with the ongoing frustration of having to cajole the machine. I felt like I was really starting to figure out how to talk to it. I was getting results.
What I really wanted to do was to see if I could create an entire add-on product using Codex in the four days I had available.
OpenAI says that if you’re a regular developer doing regular developer things (like not running hundreds of automated agents all at once), you should be able to use Codex in the $200/mo Pro plan continuously, without getting cut off.
So I decided to take the chance. I upgraded my account, paid my $200, and was back in Codex’s good graces. Could I create an entire add-on in four days?
This was a very ambitious goal for me. Without the AI’s help, I might have been able to prototype the UI look for an add-on in four days. I wouldn’t have been able to get the UI to where it was live. I’d only have maybe been able to get it to show up and look mostly right. Then I’d be out of time.
Also: The best AI for coding in 2025 (and what not to use)
I’d have needed to wait three or four months until I had another gap in time, spend the first day getting back up to speed, update all the development tools, and basically get my coding legs back. Then I might add a feature or wire the UI to save and restore its settings. Then I’d be out of time again.
I’d have needed to wait three or so months again. Rinse, wash, repeat. Net-net-net, it would usually take about a year of tiny little coding sprints to get an add-on done. Historically, that’s how it had always been for me on this type of project.
So, the idea that I might be able to do one whole add-on in four days was insanely aggressive, but I wanted to try.
Day 1: Starting (and finishing) the site analysis tool
In my previous time-limited sprint with ChatGPT Plus and Codex, I had tried the recommended best practice of providing a full product requirements document and spec to the AI. This failed miserably. I’ll talk about that in another article, where I’ll share some lessons learned. What this meant, though, was that I needed a new approach.
I decided to feed the AI assignments piecemeal. I gave it a fairly narrow instruction, worked through getting that to a workable point, and moved on. I like to start by building the UI, and then later add all the business logic, so that’s what I did.
Also: How people actually use ChatGPT vs Claude – and what the differences tell us
I expected the prototype UI development to take me a day or two. In practice, it was completed within about four hours. I gave it prompt after prompt, waiting 5-15 minutes between each, and then gave it very detailed instructions on how to tune things, piece by piece.
After that, I had it build up a table in the UI to keep track of the site data instances to let the user see what’s happening. That took another few hours.
Finally, I had it work through the business logic and build in the data acquisition catch points by linking into some provided hooks in my core security software.
By the end of the day, I had a fully-working premium add-on that transforms a private WordPress site into a security and analytics hub by logging, analyzing, and reporting all visitor activity.
Features include:
- The ability to track all login attempts (success, failure, logout) with timestamps and user details.
- The capability to detect bots and AI crawlers, and log their agents, IPs, and access patterns.
- The ability to consolidate events into a searchable, filterable, and exportable activity log.
- An analytics feature that flags brute-force attempts, suspicious logins, and traffic spikes.
- Admin-friendly reports with clear badges, summaries, and visual indicators.
- Designed for performance, security, and privacy, with all data stored locally under admin control.
To be clear. I built that in a day. Now, the AI couldn’t have done it on its own. It needed my skills as guidance because it often went down the wrong rabbit hole. Together, we turned out a very powerful product. In a day.
Day 2: The AI scraping mitigation tool
I expected to spend all four days and barely get the site tracking tool done. Instead, it was complete on the first day. So I decided to push it. Could I get another add-on done in another day?
I used the same process, with building out the UI iteratively first, then wiring the UI to save and restore settings, and then connecting the business logic. In the case of this next tool, I used the inherent site protection features that exist in my security product and made them available to protect in a new way via the add-on.
Also: How web scraping actually works – and why AI changes everything
By dinner time, I had completed another premium WordPress add-on that protects website content from AI crawlers and enables controlled licensing, blocking, and enforcement options.
Features for this add-on include:
- Integrates Really Simple Licensing (RSL) to declare how AI may use your content. My ZDNET colleague Steven Vaughan-Nichols wrote about RSL last week, and I decided to see if I could make it work in my code.
- Adds NoAI and NoImageAI meta tags and headers to block AI training on text and images.
- Provides a curated checklist to block specific AI bots via robots.txt.
- Includes an active countermeasures firewall using my security software mitigation infrastructure that returns 403 errors to banned bots.
- Offers user-friendly setup with checkboxes, templates, and status indicators.
Most protocols that are designed to discourage web scrapers, including the new RSL standard, rely on voluntary compliance from the AI software. I included those flags, but I also added the ability for users to actively block access, so if a scraper isn’t behaving itself, it won’t get in.
That was Thursday. All of that. Thursday.
Day 3: The IP blocking tool
Next up, I wanted to add IP blocking to my security product. With the help of Codex, I built a premium WordPress add-on that empowers site owners to block specific visitors or entire networks by IP, delivering stronger privacy and access control.
Features for this beastie include:
- Blocking individual IP addresses or entire ranges, with full IPv4 and IPv6 support.
- Enforcing complete site lockouts or front-end-only restrictions, depending on your needs.
- Detecting real client IPs even behind proxies or CDNs, with a Trust Proxy option.
- Managing blocks through a user-friendly interface with bulk add, CSV import/export, and search tools.
- Ensuring fast and reliable enforcement with optimized lookups, caching, and resource-saving logic.
And it’s fast. Really, really fast. There’s also a display at the top of the UI that shows how many IPs are being blocked, both using IPv4 and IPv6.
Because you can set ranges in IPv6 that boggle the mind, I actually programmed the summary display to use exponential notation for the very, very big numbers. The geek in me loves that feature.
So, that was Friday. Three products. Three days.
Day 4: The guest access control tool
Saturday was my last day for programming. I had a bunch of Monday deadlines. I would need Sunday to prepare those. So my big programming sprint had to end at dinner time Saturday night. I almost didn’t make it because we had a minor family emergency in the middle of the day. But it resolved, and I was able to continue with my coding.
This last tool is a premium add-on that lets you securely share temporary, controlled access to your private WordPress site without creating accounts or compromising privacy. Features include:
- Creating secure, random token-based guest links that bypass the login screen.
- Setting time-limited or one-time-use access, ensuring privacy reverts automatically.
- Offering frictionless entry with no accounts, usernames, or passwords required.
- Managing and instantly revoking invites from a dedicated WordPress dashboard with usage tracking.
This is a feature that my users have asked for consistently, but it was something that required a lot of very careful details and UI elements. I just never got around to it until now.
So yeah. That happened.
Okay, so let’s do the math. Over the course of ten years, I’ve managed to create 10 add-ons. That’s one add-on per year.
Last week, I created four add-ons in four days.
If I previously could only do one add-on a year, and I’m now able to do an add-on in a day with the help of Codex, you can see how I get to the metric of four years of add-ons in four days.
ChatGPT Pro did not cut me off during this programming sprint. Across the four days, I put in full work days with almost constant Codex use. It didn’t flag or throttle me even once.
I also need to point out that just because the development is complete, that’s not all there is to getting these things out as a product. Still to do (and it will take considerably more than a day each) is:
- Careful functional testing
- Creating a demo/training video
- Writing a technical white paper for folks who want deeper understanding
- Creating a product page in the online store
- Creating graphics representing the products for the add-ons list in the product
- Sending notices to users to let them know
- Figuring out some marketing plan and carrying it out
- Possibly doing some press promotion
In other words, Codex helped with the coding, but all the product marketing is my responsibility. It’s odd and a little disconcerting that the promotional prep and administrivia will take longer than the actual product development, but it’s also very cool.
From my personal productivity standpoint, it’s a good balance. I do need blocks of time allocated for programming, and those blocks are hard to come by. But all these marketing activities can be done in little chunks, when I get a half hour here or there. They don’t require the context switch that coding normally does.
For me, that means the AI has collapsed the context switch friction that has historically caused these add-ons to each take a year to develop. It appears that any time I have a free weekend, I can probably completely develop one add-on or another.
Is that worth $200 on the months I want to do this kind of coding? Probably.
What do you think? Could an AI tool really collapse years of product development into a handful of days, or is that just an edge case? Have you found that context switching slows your own productivity as much as it did mine? Would paying for extra AI access be worth it if it meant turning long, stop-start projects into weekend sprints? And more broadly, do you see AI as removing bottlenecks in your workflow or just shifting them somewhere else? Let us know in the comments below.
You can follow my day-to-day project updates on social media. Be sure to subscribe to my weekly update newsletter, and follow me on Twitter/X at @DavidGewirtz, on Facebook at Facebook.com/DavidGewirtz, on Instagram at Instagram.com/DavidGewirtz, on Bluesky at @DavidGewirtz.com, and on YouTube at YouTube.com/DavidGewirtzTV.
Leave a Reply