Easter weekend has come and gone in a blitz, and seemingly, so has this challenge. As a reminder for those coming here for the first time:
In Christian tradition, Lent is a 40-day fast and repentance preceding Easter, beginning on Ash Wednesday and ending before Maundy Thursday or Easter. In that same vein, I chose to embark on a 40-day fast from any AI use.
This was, no doubt, an interesting self-imposed challenge, and I was happy to see people from different corners of the internet stumble upon it.
It was also a challenge, that I’ve failed. I’ve debated whether or not to completely scrub this from my blog, but ultimately decided that I would not contribute to the File Drawer problem within research, and publish my findings regardless of their result.
In the days leading up to Easter, I really dreaded writing this blog post, because I felt the guilt of facing all the people that followed this journey and shamefully admitting that I couldn’t keep up the contract.
Vainly, I thought of many excuses that would still make me look good and not lose face, but I’ll be honest: I failed this challenge, I took the easy happy path, and using AI feels good.
… or does it?
Gentlemen, a short look back to the past…
Setting the ground rules for this challenge, I mentioned that I will take copious surveys around a couple of metrics to judge how this challenge is impacting my work output and satisfaction. These metrics were Satisfaction with Software Engineering as a craft, Fatigue (cognitive load), Number of PRs merged, and Confidence in released features.
This is by no means an empirical result, because it relies of my sparse self-reporting, but there are some interesting findings, even within this sparse data:
| D1 | D2 | D3 | D6 | D7 | D8-26 | D26-D40 | |
|---|---|---|---|---|---|---|---|
| Satisfaction | 6 | 4 | 8 | 6 | 9 | 3 | 2 |
| Fatigue | 8 | 8 | 3 | 9 | 8 | 8 | 9 |
| PRs Merged | 0 | 1 | 0 | 1 | 1 | 9 | 6 |
| Confidence | 8 | 9 | 8 | . | 8 | 4 | 5 |
The challenge started strong, and I was suprised at the headwinds I was facing when trying to get any meaningful work done without Claude Code. It was mental pain, but it was fun.
At around the 13-14 day mark, the challenge completely fell apart. Due to workload, I leaned back onto agentic coding to fill in the gap. I thought I could get away with peppered use of AI agents here and there and then go back to the challenge once my workload reduced. This wasn’t the case, and even minuscule AI use caused a slippery slope back into the habit of coding ‘agentically’ 100% of the time.
The silver lining about this, however, is that this relapse (lmao) means we can contrast my experience with and without AI.
Metric No AI With AI Δ Verdict
─────────────────────────────────────────────────────────────────────────────────
Satisfaction 6.6 3.0 -3.6 ▼ Worse
Fatigue 7.2 8.0 +0.8 ▲ Worse (marginally)
PRs/period 0.4 9.0 +8.6 ▲ Better
Confidence 8.2 4.0 -4.2 ▼ Worse
Verbalising this data, we can pretty much say:
I delivered more work that I was less confident about, making me more miserable in the process
Looking at each of these metrics individually:
AI is making me more productive
I am closing more tickets, I am working at a faster rate, and while I haven’t tracked this, it doesn’t seem like I’m introducing more regressions or new bugs (for now).
AI is making me hate Software Engineering
This industrialisation of my SWE tasks is draining any sort of joy I’ve had for the craft. If Agentic coding fails, it’s incredibly frustrating to untangle and debug, and even if it succeeds, finishing a task via Claude Code leaves me… quite frankly… numb. There’s no dopamine reward at the end, since I’ve outsourced the squeezing of my brain-sponge for brain-juices to Anthropic.
AI is increasing my impostor syndrome
I’ve always struggled to have confidence in the code I wrote. I would only have confidence in the patterns I’d use if I saw them used elsewhere. AI kinda supercharged this impostor syndrome, because I can always reach for Claude to interrogate the “Bell Curve of code implementations”.
When I wrote code by hand, I was more confident in it, and my abilities to pick the correct patterns.
Fatigue quantitatively unchanged, qualitatively different
Initial fatigue when starting this challenge was related to having to re-activate my atrophied coding neural pathways, fatigue towards the end of this challenge was more related to context switching.
Conclusion
This challenge was quite intense, and even emotionally so. I expected a very straightforward progression from struggling to clean my mind’s cobwebs initially, a period of frustration and ultimately, deliverance. What I found was frustration, guilt, yielding to temptation, disappointment in self.
Even though I’ve technically failed this 40 day AI fast, I think it exposed stuff about myself that I wasn’t really taking seriously, namely how closely intertwined my self-image is to programming, and the learned helplessness that these tools imbue into their users.
Starting this challenge, I was sure that I would emerge from it and declare myself saved from the bondage of AI. I was sure that I would renounce the use of AI tools for good, and proudly proclaim my newfound wisdom and virtue to the world. Now, I’m not so sure.