DEV Community

강해수 profile picture

강해수

404 bio not found

Joined on 
1024-token RAG chunks cut my storage cost in half — and nearly doubled my Claude bill

1024-token RAG chunks cut my storage cost in half — and nearly doubled my Claude bill

1
2 min read
My Durable Object processed 4 req/s instead of 40 — the culprit wasn't storage

My Durable Object processed 4 req/s instead of 40 — the culprit wasn't storage

1
2 min read
Adding one field to Notion cost me 2.5 hours. The same change in Tana took 30 seconds.

Adding one field to Notion cost me 2.5 hours. The same change in Tana took 30 seconds.

1
2 min read
My best-looking ROAS campaigns were quietly destroying subscription revenue

My best-looking ROAS campaigns were quietly destroying subscription revenue

1
2 min read
My agent dry-ran fine in staging 100 times — then wrecked production on the first real run

My agent dry-ran fine in staging 100 times — then wrecked production on the first real run

2 min read
`wrangler dev --remote` silently writes to your production KV namespace — here's the fix

`wrangler dev --remote` silently writes to your production KV namespace — here's the fix

2 min read
I audited 340 reading captures. Only 20% ever became knowledge I actually used.

I audited 340 reading captures. Only 20% ever became knowledge I actually used.

2 min read
Adding more Claude subagents made my pipeline slower — here's the specific reason why

Adding more Claude subagents made my pipeline slower — here's the specific reason why

1
2 min read
56% of my notes died before I ever had a chance to retrieve them

56% of my notes died before I ever had a chance to retrieve them

1
2 min read
My Anthropic bill dropped from $312 to $156 after I added two bash hooks to Claude Code

My Anthropic bill dropped from $312 to $156 after I added two bash hooks to Claude Code

2 min read
800 simultaneous Workers, one cache miss, $40/mo surprise — the Cloudflare coalescing fix

800 simultaneous Workers, one cache miss, $40/mo surprise — the Cloudflare coalescing fix

2 min read
73% of my context switches came from one mistake: sharing a workspace across two businesses

73% of my context switches came from one mistake: sharing a workspace across two businesses

1
2 min read
Naver Powerlink quality score: 11 of 17 ad groups flipped from medium high — here's what actually moved them

Naver Powerlink quality score: 11 of 17 ad groups flipped from medium high — here's what actually moved them

2 min read
D1 Schema Migrations with AI Agents: The DDL-in-Transaction Trap That Kills Zero-Downtime Deploys

D1 Schema Migrations with AI Agents: The DDL-in-Transaction Trap That Kills Zero-Downtime Deploys

2
2 min read
I tagged 34 decision notes in Obsidian over 18 months. Reading them back broke my Notion setup.

I tagged 34 decision notes in Obsidian over 18 months. Reading them back broke my Notion setup.

1
2 min read
Adding one field to our CAPI payload moved match rate from 66% to 79% overnight

Adding one field to our CAPI payload moved match rate from 66% to 79% overnight

1
2 min read
A one-line cache key bug cost me $187/month and leaked advertiser data across tenants

A one-line cache key bug cost me $187/month and leaked advertiser data across tenants

1
2 min read
My Durable Objects alarm loop burned CPU for 3 days before I noticed — here's what the docs miss

My Durable Objects alarm loop burned CPU for 3 days before I noticed — here's what the docs miss

2 min read
ASC burned ₩3M in 72 hours on out-of-stock SKUs — here's the structural fix

ASC burned ₩3M in 72 hours on out-of-stock SKUs — here's the structural fix

2 min read
Cutting Claude API Costs in Half with a 3-Tier Routing System (Haiku/Sonnet/Opus)

Cutting Claude API Costs in Half with a 3-Tier Routing System (Haiku/Sonnet/Opus)

1
2 min read
FTS5 on Cloudflare D1 cut my Korean keyword search from 220ms to 22ms — here's what almost broke it

FTS5 on Cloudflare D1 cut my Korean keyword search from 220ms to 22ms — here's what almost broke it

1
2 min read
tCPA thrashed for 35 days straight — then beat manual CPC by 35% on cold prospecting

tCPA thrashed for 35 days straight — then beat manual CPC by 35% on cold prospecting

2 min read
60% of My $312 Anthropic Bill Came From One Missing Pattern: Compensating Actions

60% of My $312 Anthropic Bill Came From One Missing Pattern: Compensating Actions

1
1
2 min read
KV locks for Workers cron jobs will silently fail — here's what I use instead

KV locks for Workers cron jobs will silently fail — here's what I use instead

2 min read
My 7-step prompt chain failed silently at step 6. Here's what replaced it.

My 7-step prompt chain failed silently at step 6. Here's what replaced it.

1
2 min read
1,400 rows broke my D1 batch — here's why 7 statements beat 1,400

1,400 rows broke my D1 batch — here's why 7 statements beat 1,400

1
2 min read
31% of my week was spent figuring out what I was supposed to be doing

31% of my week was spent figuring out what I was supposed to be doing

2 min read
Funnel-stage ad structure collapses under budget cuts. Product-line holds longer — here's the failure mode for each.

Funnel-stage ad structure collapses under budget cuts. Product-line holds longer — here's the failure mode for each.

2 min read
60% of My $312 Anthropic Bill Came From One Silent Loop — Here's How I Found It

60% of My $312 Anthropic Bill Came From One Silent Loop — Here's How I Found It

1
1
2 min read
400 ad accounts 3 subrequests = silent data loss in Cloudflare Workers (and how I fixed it)

400 ad accounts 3 subrequests = silent data loss in Cloudflare Workers (and how I fixed it)

2 min read
I lost a client note to iCloud sync conflicts — here's the Obsidian mobile workflow I rebuilt after

I lost a client note to iCloud sync conflicts — here's the Obsidian mobile workflow I rebuilt after

2 min read
The Meta exclusion layer most accounts skip is costing them 8-15% of lookalike impressions

The Meta exclusion layer most accounts skip is costing them 8-15% of lookalike impressions

2 min read
I let Claude Code run --dangerously-skip-permissions on my production DB. Here's what I changed.

I let Claude Code run --dangerously-skip-permissions on my production DB. Here's what I changed.

2
1
2 min read
I tracked every async writing block for 90 days. Output nearly doubled — but not how I expected.

I tracked every async writing block for 90 days. Output nearly doubled — but not how I expected.

2 min read
The NOT NULL trap that killed my Cloudflare Worker at 2:47am

The NOT NULL trap that killed my Cloudflare Worker at 2:47am

2 min read
My second brain cost me 140+ hours before it saved me anything

My second brain cost me 140+ hours before it saved me anything

2 min read
Naver Brand Search shows 5 –9 LTV:CAC and it's the least useful number on my dashboard

Naver Brand Search shows 5 –9 LTV:CAC and it's the least useful number on my dashboard

2 min read
More parallel subagents made my pipeline slower. Here's the data.

More parallel subagents made my pipeline slower. Here's the data.

2
1
2 min read
Your R2 presigned URL has a shorter TTL than you think — here's why

Your R2 presigned URL has a shorter TTL than you think — here's why

1
2 min read
I tracked every context switch across two businesses for 90 days. The 23-minute recovery rule is misleading.

I tracked every context switch across two businesses for 90 days. The 23-minute recovery rule is misleading.

1
1
2 min read
40% of my lookalike rebuilds were unnecessary — here's the signal-triggered system I use instead

40% of my lookalike rebuilds were unnecessary — here's the signal-triggered system I use instead

2 min read
My agent was 'succeeding' on Slack while silently doing nothing — here's the monitoring stack that caught it

My agent was 'succeeding' on Slack while silently doing nothing — here's the monitoring stack that caught it

2
2 min read
D1 read replicas returned stale data for 6 seconds — here's what I measured and how I designed around it

D1 read replicas returned stale data for 6 seconds — here's what I measured and how I designed around it

2 min read
I tracked AI rewrite rates for 12 weeks. Notion AI needed fixing 70% of the time. Claude: 30%.

I tracked AI rewrite rates for 12 weeks. Notion AI needed fixing 70% of the time. Claude: 30%.

2 min read
25–35% of our best customers were aging out of retargeting before we ever showed them an ad

25–35% of our best customers were aging out of retargeting before we ever showed them an ad

2 min read
I cut my Claude system prompt from 4,200 to 820 tokens to save money. It broke production in 3 days.

I cut my Claude system prompt from 4,200 to 820 tokens to save money. It broke production in 3 days.

2 min read
My weekly review clocked 14 minutes median — here's the one structural change that made it stick

My weekly review clocked 14 minutes median — here's the one structural change that made it stick

1
2 min read
A Merchant Center disapproval wiped 40% of our SKUs the day a 6-week promo launched

A Merchant Center disapproval wiped 40% of our SKUs the day a 6-week promo launched

2 min read
I wasted $43 rebuilding a Vectorize index the wrong way — here's the $5.50 fix

I wasted $43 rebuilding a Vectorize index the wrong way — here's the $5.50 fix

2 min read
I tracked 34 'dead opens' in Obsidian over 3 weeks. That's why I switched to Tana.

I tracked 34 'dead opens' in Obsidian over 3 weeks. That's why I switched to Tana.

2 min read
Naver catalog quality score: the field that tanked impression share wasn't the one we were optimizing

Naver catalog quality score: the field that tanked impression share wasn't the one we were optimizing

2 min read
A Durable Object I thought was idle cost me $40/mo — here's exactly why

A Durable Object I thought was idle cost me $40/mo — here's exactly why

2 min read
loading...