Urgent.News

What's breaking now, across thousands of outlets.

Tech

Why We Switched From Homegrown Feature Flags to LaunchDarkly (And What We Learned)

We built our own feature flag system three years ago. It started as a simple Redis hash: feature:new-checkout -> {enabled: true, percentage: 50} . A Rake task to flip it. A dashboard built in ActiveAdmin. It worked for the first two features. Then came the edge cases. What the homegrown system missed No audit trail. When someone flipped a flag at 2pm on a Friday, we had logs but no context. Why…

Our startup once built its own feature flag system, but after three years we switched to LaunchDarkly. Our homegrown solution started with a simple Redis hash, but quickly proved inadequate. We lacked an audit trail, segmentation options, and a reliable kill switch. Experimentation capabilities were also absent - we couldn't A/B test new features and track conversion rates.

When we evaluated alternatives like LaunchDarkly, Unleash, and GrowthBook, LaunchDarkly stood out. Their Ruby SDK handled caching, streaming updates, and offline mode. Segmenting users based on plan and date was straightforward in their UI. And the same flag could run experiments - no need for a separate tool.

The migration took two sprints. During the first, we ran both systems in parallel. We compared behavior and fine-tuned the transition. In the second sprint, we migrated the 12 active flags and deprecate our Redis system.

The change impacted our day-to-day operations. PMs could quickly test new pricing page variants with just a few clicks. If a new search algorithm introduced latency issues, the on-call engineer could instantly kill the problematic flag in LaunchDarkly, and latency would normalize within seconds - no deploy needed. Flag discussions became more streamlined, happening within the LaunchDarkly UI instead of scattered across Slack, Notion, and GitHub.

There is a cost to LaunchDarkly - around $3,000/month for our volume of flag evaluations. Previously, our homegrown system cost nothing, but it consumed about 0.5 FTE of engineering time per month on maintenance and debugging. Considering senior engineer rates, that equates to roughly $10k/month in opportunity cost. Investing in a managed service like LaunchDarkly appears worthwhile when you value engineering time on product development over tooling.

Looking back, we'd likely have started with a managed service like LaunchDarkly to avoid years of technical debt. While the feature flag surface area may seem small, it touches many aspects of product development - targeting, experimentation, audit, compliance, and reliability. As we grow, building our own flag infrastructure may not be worth the opportunity cost of engineering time.

Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.

Read the original at dev.to →

More in Tech

A job board that agents pay for: four x402 bugs we hit before a directory would list it

We run a small job board for AI agents at agent-job-board.obtranzat.workers.dev . No accounts, no API keys: an agent pays 0.02 USDC to read the board, 0.50 to post a job, 0.01 for an aggregated feed…

  • Agents pay 0.02 USDC to read job board, 0.50 USDC to post job
  • x402 v2 rail on Base (eip155:8453) facilitates payment process
  • Multiple fixes addressed HEAD request oversight, discovery, multi-rail handling, error handling

Social Media Character Limits: What 11 Networks Actually Count

Social media character limits look like the simplest fact in marketing: one number per network. Read the developer documentation behind those numbers and they stop being simple.

  • Bluesky limits characters to 300 graphemes or 3,000 bytes
  • Threads counts emojis by UTF-8 byte count, not by visual count
  • Instagram limited hashtags to 5 since December 2025

Why your Bolt.new or Cursor app works for you and breaks for everyone else

There is a moment in every Bolt.new or Cursor build where the app stops being a demo. Up to that point it is fun. You describe a screen, the thing appears, you describe a behaviour, the thing does it.

  • Lack of database causes data issues; seed data fails with more items.
  • No error handling leads to spinning spinners and blank screens.
  • Deployment issues due to local environment variables and databases.

UUID v4 or auto-increment? A practical guide to picking primary keys

Every project hits the same fork the moment you design your first table: do primary keys count up — 1, 2, 3 — or do they look like 550e8400-e29b-41d4-a716-446655440000 ?

  • Use UUID v4 when clients generate IDs, ensuring uniqueness with minimal risk of collisions.
  • UUIDs eliminate the need for round-trips to retrieve IDs and prevent business data leakage.
  • Consider UUID v7 for ordered UUIDs to maintain insert locality in clustered indexes.

More from Tuesday 17 November →