DEV Community: Andrey Andrade

The time bomb in your codebase (and mine too)

Andrey Andrade — Sun, 22 Feb 2026 18:07:35 +0000

Here's something it took me years to accept: no matter how good you are, your codebase is rotting right now.

It's not your fault. It's biology.

The number that explains everything

Your brain can hold 4 to 7 chunks of information in working memory at once. That's the same for me, for you, for the most brilliant senior dev you've ever worked with. It's a biological limit, not a skill issue.

Now think: how many implicit interactions does the codebase you maintain have? 50 thousand lines? 200 thousand? Half a million?

You're trying to monitor a system with thousands of invisible dependencies using a processor that runs 7 threads. It won't work.

The cycle I've seen dozens of times

I've been working with software for 22 years. The pattern is always the same:

Month 1 — The architect designs everything with clarity. Decisions make sense. Everyone's aligned.

Month 6 — The architect is still there, but no longer remembers why they chose that 30-second timeout instead of 5. "There must have been a reason."

Month 12 — Two of the three original engineers left. The reason for the timeout? It left with them. What remained was a comment // don't change this with no explanation.

Month 18 — The new dev, who is competent, changes the timeout to 5s because "30s is absurd." Production breaks at 3 AM. Nobody understands why.

The problem is never the new dev. The problem is that context doesn't survive time.

The failures that kept me up at night

After two decades, I collect these stories:

The innocent hook. A usePopup() that adds a global listener to the DOM. Clean, tested, code review approved. Except it was used in 100 components. Now there are 100 listeners firing on every click anywhere on the page. No individual code review will catch this — because each review only sees its one usage.

The phantom cache. A legitimate refactor changes how an object is constructed. Functionally identical. Referentially different. The cache that depended on === never works again. Tests pass. App doesn't crash. It just gets silently slower, week after week, until someone opens a ticket three months later.

The elegant N+1. A loop with an ORM that produces beautiful code to read and 50 separate SQL queries to execute. The dev sees elegance. The database sees a bombardment.

The pattern is always the same: the problem isn't visible at the abstraction level you're working at. You'd need to see multiple levels simultaneously — and your 4 to 7 memory chunks won't let you.

Where AI actually changes the game

I know there's a lot of hype. I know half the promises about AI in software are marketing. But there's one thing AI does that is structurally different from any tool that's ever existed:

Giant context window. When you change a line, AI can see the 47 modules that call that function, check if any of them depend on referential equality, and detect if your "simple fix" will silently break something on the other side of the codebase. Not because it's smarter — because it has more active working memory.

Tireless vigilance. Friday at 5 PM, deploy scheduled for Monday, code review gets more superficial. On the 100th PR of the week, attention is lower than on the first. AI applies the same scrutiny to PR #1 and PR #100. No "this time is different," no "it's just a hotfix."

Institutional memory. "Why does this service have a 30s timeout?" — the new dev doesn't know, the 2-year-old Slack thread doesn't show up in search. An AI with access to history remembers: "INC-4521, March 2024, spot instances with cold start up to 25s."

But hold on — humans are still irreplaceable

Before anyone thinks I'm saying AI will replace devs: it won't.

AI is extraordinary at following patterns and detecting violations. But it doesn't invent new paradigms. It doesn't know you're accepting that tech debt because the market window closes in 3 weeks. It doesn't know "don't deploy on Monday because the payments team deploys on Tuesday." It doesn't know what happened during last Black Friday.

Judgment, creativity, business context, tribal knowledge — that's still ours.

The real partnership

The future isn't "devs vs. AI." It's:

AI handles consistency and scale — detects violations, maintains memory, checks cascade impacts, watches without fatigue
Humans focus on judgment and creativity — decide what to build, make strategic trade-offs, invent new things, navigate organizational politics

Teams that resist this out of pride ("I can review everything myself") will accumulate entropy faster. Not because they're worse devs — because they're humans with fixed cognitive limits facing growing complexity.

The clock is ticking

If your codebase is over 18 months old, the entropy is already there. The question isn't whether it exists — it's whether you have a system capable of continuously defusing it.

Or whether you're counting on 4-7 chunks of human memory to monitor hundreds of thousands of lines.

I wrote an expanded version with diagrams, charts, and detailed technical examples here: stickybit.com.br/codebase-timebomb/

If you maintain a legacy codebase (and who doesn't?), it's worth a read.

I scraped 1,600 DEV.TO articles from 2025 and built an infographic

Andrey Andrade — Sun, 11 Jan 2026 22:05:10 +0000

Last weekend I got curious: what's actually trending on DEV.TO in 2025?

So I wrote a Go script, hit the API, and pulled 1,600 featured articles. Then I built an infographic to visualize the data.

Here's what I found:

AI is everywhere

The #ai tag appears in 33% of all articles. Add related tags like #machinelearning, #llm, and #agents, and you're looking at 40%+ of content being AI-related.

2025 is definitely the year AI took over the dev conversation.

The numbers

1,600 articles
54,716 reactions
19,543 comments
7,084 minutes of reading time (that's ~5 days straight)

Average article: 4.4 min read, 12 comments, 34 reactions.

December was crazy

336 articles in December alone — 47% more than November. Year-end retrospectives and 2026 planning posts everywhere.

April was the quietest month with only 71 articles.

Top tags after AI

ai (530)
webdev (397)
programming (287)
discuss (244)
javascript (152)

JavaScript still holding strong, but AI content is 3.5x more common.

The infographic

I put everything into an interactive infographic with animated charts, tag clouds, and top articles by engagement.

Check it out: devto-2025/

Built with Go backend and vanilla JS + Tailwind. No frameworks, no npm install, just straightforward code.

The data was collected via the official DEV.TO API (Forem). Only featured articles from 2025 were included.

Curious about anything specific in the data? Drop a comment.

I Analyzed 4.78M HN Items to Create an Interactive 2025 Infographic

Andrey Andrade — Fri, 02 Jan 2026 18:46:38 +0000

I downloaded the complete Hacker News archive (160 SQLite shards via StaticNews) and analyzed all 2025 data to create a GitHub Octoverse-style interactive infographic.

https://andreyandrade.com/static/hn-universe/

Key Findings

AI Dominated 2025

The AI conversation on HN shifted dramatically:

AI-related stories grew from 8% to 13% of all submissions
Claude overtook ChatGPT in mentions during H2
DeepSeek was January's phenomenon, then vanished (-92%)
The "Models Race" shows Claude at +482% growth vs GPT's +65%

Programming Languages

Language	Growth	Avg Points
Zig	+282%	59
Rust	+97%	45
Ruby	+41%	38
Go	+23%	42

Zig is the most hyped language of 2025. Ruby had a surprising comeback.

Job Market (4,885 postings analyzed)

Remote: 57% ████████████████████░░░░░░░░░░░░░░░
Onsite: 23% ████████░░░░░░░░░░░░░░░░░░░░░░░░░░░
Hybrid: 20% ███████░░░░░░░░░░░░░░░░░░░░░░░░░░░░

71% mention AI/ML skills
Only 3% are junior positions
Average salary: $180k (when disclosed)

Engagement Patterns

Some surprising discoveries:

"Tell HN" posts average 5x more points than "Show HN"
Best posting time: 15:00 UTC (11am EST)
September had the most viral posts (34 stories with >1000 points)
Weekend posts get 23% less engagement

The Stack

Data: https://static.hn (160 SQLite shards, 4.78M items)
Analysis: Go (processes all shards in parallel)
Visualization: Vanilla HTML/CSS/JS with CSS animations

The analyzers are written in Go and process all 160 shards concurrently. Each shard contains ~30k items.

How I Built It

Downloaded all 160 shards (~2GB total)
Wrote Go scripts to query patterns across shards
Aggregated temporal data (monthly trends, hourly patterns)
Built the infographic with pure CSS animations (no JS frameworks)

The trickiest part was normalizing the data - HN's API changed formats over the years, and some items have missing fields.

What Surprised Me Most

Tell HN vs Show HN: I expected Show HN (project launches) to get more engagement than Tell HN (personal stories). It's the opposite - Tell HN averages 54.7 points vs Show HN's 11.2. The community values personal narratives more than product launches.

The DeepSeek Spike: In January 2025, DeepSeek dominated discussions. By March, mentions dropped 92%. The AI hype cycle moves fast.

Junior Jobs Scarcity: Only 3% of "Who is Hiring" posts target juniors. The industry's entry barrier keeps rising.

Check It Out

Andrey Andrade

The infographic has smooth scroll animations and works on mobile. All data is from real HN submissions and comments.

What patterns have you noticed on HN this year?

Built by https://andreyandrade.com - Backend engineer focused on Go, data analysis, and AI integration.

Dados Abertos de Cidades Brasileiras em JSON - DDDs, Bairros, Estados

Andrey Andrade — Sat, 20 Dec 2025 20:19:20 +0000

🇧🇷 Dados Abertos de Cidades Brasileiras

Disponibilizei um repositório com dados estruturados sobre cidades brasileiras, pronto para usar em qualquer projeto.

## O que está incluído

| Arquivo | Descrição |
| cidades.json | 20 maiores cidades com população, DDD, região

| estados.json | 27 estados + DF com capitais |
| ddds.json | 67 DDDs com cidades associadas |
| bairros/*.json | ~450 bairros de 20 cidades |

## Uso Rápido

Casos de Uso

Formulários: Validar DDDs, autocompletar cidades
Dashboards: Dados demográficos por região
Apps: Listar bairros de uma cidade
Data Science: Análises geográficas

Link

🔗 GitHub: https://github.com/andradeandrey/autocidade-dados-abertos

📊 Fonte: https://autocidade.com

Contribuições são bem-vindas! Se quiser adicionar mais cidades ou dados, abre um PR.

Tags: #opensource #brasil #json #api #dados