A new study is making the rounds: 2 weeks without smartphone internet significantly improved sustained attention — effects similar to being a decade younger. Bryan Johnson shared it. It went viral. 800K+ views in hours.
But here's what nobody in the replies is connecting: your distracted brain is directly inflating your AI development costs.
I've been building two Mac apps as a solo dev, and after tracking my AI spending and my focus patterns side by side for 90 days, the correlation is undeniable.
The Numbers Nobody Talks About
On days I rated as "focused" (minimal feed browsing, deep work blocks):
- Average AI spend: $4.20/day
- Average prompts: 23
- Retry rate: 8%
On days I rated as "scattered" (checking feeds, context-switching):
- Average AI spend: $11.80/day
- Average prompts: 47
- Retry rate: 34%
Same projects. Same tools. Same models. Nearly 3x the cost.
Why Distraction Makes AI Coding Expensive
1. The Retry Tax
When you're distracted, you write worse prompts. Worse prompts get worse results. Worse results mean more retries. Each retry burns tokens.
A focused developer writes one clear, well-scoped prompt. A distracted developer writes three vague ones and then a fourth to fix what the third broke.
At $0.015-0.06 per 1K tokens on Opus, those retries add up fast.
2. The Context Window Bloat
Distracted developers don't scope their requests. Instead of "refactor the authentication middleware in auth.ts," they say "fix the login."
The AI tool then pulls in everything it thinks might be relevant — your entire project structure, unrelated files, test suites. Your context window balloons from 2K tokens to 40K tokens. That's a 20x cost multiplier on a single request.
3. The Model Mismatch
When you're focused, you think about what you're asking. "Does this need Opus-level reasoning, or is Sonnet fine?"
When you're distracted, everything goes to your default model. I found that 73% of my Claude spend was on tasks that Haiku handles identically. Not similarly — identically. Same output, 50x less cost.
4. The "Let Me Check Twitter First" Loop
This is the killer. You hit a bug. Instead of thinking through it, you "just quickly check" your feed. 20 minutes later, you come back having lost all mental context. Now you need the AI to re-explain what it was doing. More tokens. More cost. More time.
What I Actually Did About It
Step 1: Block the Feed, Not the App
I didn't delete social media. I didn't go full digital monk. I blocked the algorithmic feed on each platform while keeping everything else functional.
- Instagram: No Explore page, no Reels tab, but DMs and Stories work fine
- Twitter/X: No For You feed, but Search and notifications work
- YouTube: No homepage recommendations, but direct links and subscriptions work
- Reddit: No front page algorithm, but specific subreddits work
The key insight: the feed is the addiction mechanism, not the platform. Remove the feed and your phone becomes a tool again instead of a slot machine.
I built Monk Mode ($15, one-time) specifically because no existing blocker did this at the feed level. Every blocker I tried was all-or-nothing — block the entire site or nothing. That's like removing your car engine because you drive too fast.
Step 2: Track Per-Request AI Costs in Real Time
You can't optimize what you can't see. I put a real-time token cost counter in my menu bar using TokenBar ($5, one-time) so I could see exactly what each AI request costs as it happens.
The behavior change was immediate. When you watch $0.85 drain for a single lazy prompt that Haiku handles for $0.02, you start thinking real hard about what model you're using and how you're scoping your requests.
Step 3: Correlate the Two
After a month of tracking both focus quality and AI spend, the pattern was clear:
| Focus Rating | Avg Daily AI Spend | Avg Retries | Productive Output |
|---|---|---|---|
| High (no feeds) | $4.20 | 2-3 | 4-5 features |
| Medium (some feeds) | $7.40 | 5-6 | 2-3 features |
| Low (heavy feeds) | $11.80 | 8-10 | 1 feature |
The most expensive AI day ($23.40) was also my least productive day AND the day I spent the most time on algorithmic feeds.
The Uncomfortable Math
Let's say you're an average developer spending $150/month on AI coding tools (Claude, Cursor, Codex, etc.).
If distraction is inflating that by even 40% (which is conservative based on my data), that's $60/month in wasted tokens — directly attributable to poor focus caused by algorithmic feeds.
$60/month × 12 months = $720/year in unnecessary AI costs.
A one-time $20 investment in feed blocking + cost tracking tools pays for itself in the first week.
The Bigger Picture
The Bryan Johnson study found that 2 weeks without smartphone internet made people's attention spans a decade younger. But you don't need to go that extreme.
You just need to remove the specific thing that's hijacking your attention: the algorithmic feed.
Everything else about your phone and computer is fine. Email doesn't create compulsive behavior. Google Maps doesn't fragment your attention. Slack messages, while sometimes annoying, don't create infinite scroll loops.
It's specifically the content that's been algorithmically optimized to maximize your time-on-app. Remove that, and you get most of the attention benefits without becoming a digital hermit.
And if you're a developer using AI tools, the cost savings alone make it worth trying.
TL;DR
- A viral study shows 2 weeks without smartphone internet improves attention by a decade
- You don't need to go that extreme — just block algorithmic feeds
- Distracted developers spend 2-3x more on AI tools due to retry loops, context bloat, and model mismatch
- I tracked this for 90 days: focused days = $4.20 avg AI spend, scattered days = $11.80
- Tools I use: Monk Mode ($15) for feed-level blocking, TokenBar ($5) for real-time AI cost tracking
- Both are Mac-native, one-time purchase, no subscriptions
What's your monthly AI tool spend? And have you noticed a connection between your focus quality and your token usage? Drop it in the comments — genuinely curious if others are seeing the same pattern.
Top comments (0)