← the writing custom software & architecture · 22 essays

Custom Software & Architecture

Bespoke web development, build vs buy, microservices, real-time apps, and what custom actually costs in 2026. For founders past the no-code phase.

read the custom software & architecture guide first →

16 jul 2026 Build an AI Agent With Vercel AI SDK and Supabase I spent three days wiring up an AI agent for a client project using Vercel AI SDK and Supabase. Here's everything that worked, everything that didn't, and the exact patterns I'd use again.
16 jul 2026 Supabase vs Vercel Postgres (2026): Which One Wins? I've shipped Next.js projects on both. Supabase and Vercel Postgres are not interchangeable choices, one is a backend platform, the other is a convenience layer. Here's how to pick correctly before you're three sprints deep and regretting it.
31 may 2026 How to Modernize a Legacy Web Application in 2026 A step-by-step guide to modernizing a legacy web application in 2026: the strangler-fig pattern, replatform vs rebuild, framework migrations, and an SEO-safe cutover.
31 may 2026 Legacy Application Modernization in 2026: A Practical Playbook A practical 2026 playbook for legacy application modernization: the five strategies, when to rebuild vs replatform, realistic cost ranges, and how to migrate without losing rankings.
11 may 2026 I built a Directus admin and a sales page in 24 hours. Here is what happened. Live deploy on Railway, three dashboards, four bookmarks, five FAL-generated hero images, one Spline scene, three production bugs caught in real-time, one 2,400-word pillar page that pitches the whole thing as a service. The cost and the timeline, honestly.
10 may 2026 9 Custom CRM Advantages Over HubSpot Uncover the 9 distinct capabilities a custom CRM offers that HubSpot can't match. Learn how bespoke options can truly transform client handling.
10 may 2026 Custom CRM vs Popular Options: The Real 2026 Breakdown Wondering whether to go custom or stick with popular CRMs? Let's break down the real 2026 numbers: Custom CRM compared with HubSpot, Salesforce, and Pipedrive.
10 may 2026 Building a Custom CRM in 14 Weeks Discover Seahawk Media's practical 5-phase playbook for building a custom CRM in 14 weeks. Real-world insights and lessons from 9 years of development experience.
10 may 2026 When HubSpot Just Doesn't Cut It Anymore Discover why custom CRM development may be the right choice in 2026. When HubSpot falls short, explore real stories, numbers, and practical tools.
7 may 2026 Kimi and Minimax: how I use China's AI for deep research and design mockups Two Chinese AI tools, Kimi by Moonshot and Minimax, are doing deep research and full-app design mockups at a level I have not seen from US tools yet. How my team uses them daily alongside Claude and GPT, with the exact workflow.
7 may 2026 Rapnet vs Nivoda vs IDEX: choosing your diamond feed in 2026 Three diamond feeds, three integration paths, three pricing models. The honest comparison from someone who has shipped all three on production jewelry sites.
5 may 2026 What Claude Code actually costs at agency scale in 2026 Claude Code costs Seahawk roughly 1,200 to 2,400 USD per engineer per month at production usage in 2026. The honest breakdown of what we spend, what drives variance, and the math.
5 may 2026 CLAUDE.md for agencies: scaling AI coding standards across client projects Solo-dev CLAUDE.md advice fails at agency scale. The three-layer model (agency-wide / client-specific / per-engagement) we run at Seahawk across 200+ client repositories.
5 may 2026 The MCP servers I run in production at Seahawk: an honest stack The MCP ecosystem has 200+ servers; the production stack we actually run is 8. Filesystem, Puppeteer, Postgres, Brave, GitHub, Sentry, Notion, Linear. What we tried and dropped.
5 may 2026 Claude Code for WordPress: the workflow my agency runs in 2026 Claude Code is built for code, not WordPress, but the agency that runs it well ships at 3x velocity. The CLAUDE.md, the MCP stack, and the WordPress workflows that compress most.
3 may 2026 Agentic Development: Letting Claude Drive the Whole Loop Claude doesn't just autocomplete anymore. It plans, writes, runs tests, reads the error, and tries again, all without you babysitting it. Here's what I've learned letting it drive the whole loop.
2 may 2026 Prompt Engineering for Production Code: Hard Lessons After shipping code with AI assistance across hundreds of projects, I've learned that prompt engineering isn't a soft skill, it's the difference between production-ready output and a debugging nightmare at 11pm.
1 may 2026 Enterprise DAM in 2026: Costs, Choices, and Why It Fails Enterprise digital asset management sounds boring until you're three months into a $180k implementation and nobody can find the approved logo. Here's what DAM actually costs, how to choose, and why most projects quietly die.
30 apr 2026 Online Auction Site Tech Stack: What I'd Build in 2026 After 9 years and 12,000+ sites, here's the exact stack I'd choose to build an online auction platform in 2026, real-time bidding, payments, infra, and all the mistakes I'd avoid this time around.
30 apr 2026 Building a Real-Time Auction Site with Next.js & Supabase Real-time bidding is one of the hardest UI problems to get right. Here's exactly how I'd architect a live auction site with Next.js and Supabase, from schema design to WebSocket subscriptions, without losing your mind in the process.
30 apr 2026 Build vs Buy SaaS: A Decision Framework for Founders Every founder hits this wall eventually. Do you build the thing yourself, or pay someone else's subscription forever? Here's the framework I use after 9 years and 12,000+ sites.
30 apr 2026 How to Cost Custom Software in 2026: A Working Estimator for Founders Quoting custom software is where founders haemorrhage money, or walk away from deals that could've made them. Here's the estimator framework I actually use, with real numbers attached.