Behind the Scenes: Same Problem, Two Eras
2026-06-15
This portfolio (nelthor.qzz.io) was built using agentic development workflows.
But what does that actually mean? Let me show you by comparing how I'd build a feature in 2022 vs 2026.
The Traditional Way (2022)
Building a portfolio feature at Newfold Digital looked like:
- Ticket arrives in Jira with acceptance criteria
- Plan the implementation in a design doc
- Write code in a feature branch
- Write tests — usually after the code
- Push to Jenkins for CI
- PR review with another engineer
- Merge and deploy — maybe same day, maybe next week
- Verify in production
Each step was manual. Each handoff created delay. The bottleneck was human bandwidth.
The Agentic Way (2026)
Building the same portfolio feature with AI-assisted workflows:
- Story is refined in the backlog with Gherkin scenarios
- AI proposes an implementation plan
- I review and approve the plan (5 minutes)
- AI generates the code — components, tests, types
- I review the diff — focus on logic, not boilerplate
- Agent runs lint, typecheck, tests
- Deploy —
git push, CI handles the rest - Verify — the site updates in minutes
The bottleneck shifted from writing code to reviewing intent.
Head-to-Head
| Step | Traditional (2022) | Agentic (2026) |
|---|---|---|
| Planning | 2-4 hours | 5-10 minutes |
| Coding | 4-8 hours | AI generates, I review |
| Testing | 2-4 hours (manual) | AI generates, I validate |
| Review | 1-2 hours (sync) | Async, focused on logic |
| Deploy | Manual Jenkins trigger | Push-to-deploy |
| Total | 1-2 days | 1-2 hours |
What Changed
The work didn't disappear — it moved up the abstraction ladder:
- Before: I spent 60% of my time writing code and 40% thinking
- After: I spend 80% thinking and 20% reviewing code
The agent doesn't replace judgment. It amplifies it.
The Tools
This portfolio was built using:
- OpenCode — agent orchestration platform
- Next.js 15 — static export framework
- Tailwind CSS 4 — styling
- GitHub Pages — hosting
- Supabase — data layer (for other projects)
- Custom AI skills — project-foundation, sprint-development, product-management
Every feature followed the same cycle: plan → implement → review → deploy. The AI never worked alone. It proposed, I approved. It coded, I reviewed. It deployed, I verified.
The Real Takeaway
The same discipline that catches bugs in production also catches broken layouts, stale content, and inconsistent branding. Quality is quality, whether the code is hand-typed or AI-generated.
Agentic development is not about replacing engineers. It's about shifting the work from writing code to directing quality.