My journey to create a SaaS that would solve my content creation pain. part 2

2026-07-13 · Source: Machine Learning on Medium · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering · Depth: Intermediate, medium

Summary

The article details a personal journey to create a SaaS for content creation using AI agents, focusing on a structured development workflow to mitigate the cognitive cost of reviewing agent-generated code. It outlines a fixed project structure utilizing `description.md`, `ui-ux.md`, and `architecture.md` files, alongside a high-level architecture diagram. The author employs a Spec-Driven Development approach, using larger, more capable models for planning (generating `spec.md`, `plan.md`, `task.md`) and cheaper, smaller models like DeepSeek V4 for execution. The development environment includes a VPS, vscode-server, and an iPad, leveraging Chinese and Brazilian LLMs. While monetary costs are low, approximately US\$10-15 per month for models, the author emphasizes the significant cognitive cost of reviewing agent code, citing surveys from 2026 indicating 81% of engineering leaders reported increased review time, with 31% of a developer's day consumed by it.

Key takeaway

For AI Engineers and ML Directors adopting agentic coding, recognize that while monetary costs for LLMs are decreasing, the cognitive burden of code review is substantial. You should implement structured documentation and a tiered model workflow—using powerful models for planning and cheaper ones for execution—to manage this overhead and ensure productivity gains are realized. Be prepared for a 70-80% confidence level in personal projects, adjusting for corporate environments.

Key insights

A structured workflow using tiered LLMs and fixed documentation mitigates the high cognitive cost of reviewing agent-generated code.

Principles

• Fixed documentation structures accelerate project organization.
• Plan with large models, execute with cheaper, smaller models.
• Cognitive cost of agent code review is a significant factor.

Method

Implement a Spec-Driven Development workflow: define `description.md`, `ui-ux.md`, `architecture.md`, then generate `spec.md`, `plan.md`, `task.md`. Use large LLMs for planning, smaller ones for execution.

In practice

• Use DeepSeek V4 for cost-effective code generation.
• Set up a VPS with vscode-server for iPad development.
• Organize tasks as GitHub issues for visual tracking.

Topics

• AI Agents
• Agentic Coding
• Spec-Driven Development
• LLM Workflow Optimization
• Code Review Efficiency
• Small Language Models

Best for: AI Engineer, Software Engineer, Director of AI/ML

Related on AIssential

• The Hidden Reality of Agentic Coding: Building as a Developer on Free AI Tools in 2026 — Relying on free AI coding tools introduces significant coordination overhead and hidden productivity costs for developers.
• 🤖AI Agents Weekly: MiniMax-M2.1, GLM-4.7, MiniMax-M2.1, LaMer Meta-RL, Google's 2025 AI Breakthroughs — Effective LLM coding requires structured planning, granular tasks, comprehensive context, and rigorous human oversight.
• A Year Of Vibes — Agentic coding tools are fundamentally reshaping software engineering workflows, challenging traditional practices and human-machine relationships.
• You NEED to try these open-source AI projects RIGHT NOW — Open-source AI projects offer diverse solutions for enhanced search, local document processing, agentic engineering, and LLM cost reduction.
• The Debate Over Anthropic’s New Product: Price or Existential Dread? — AI agent pricing and capabilities are sparking an existential re-evaluation of traditional software development workflows and costs.
• Compiling Agentic Workflows into LLM Weights: Near-Frontier Quality at Two Orders of Magnitude Less Cost — Compiling agentic workflows into LLM weights creates efficient, cost-effective "subterranean agents" for procedural tasks.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Machine Learning on Medium.