Articles, reflections, and tutorials on software engineering and technology.
How I designed Crypto Trader, a multi-agent system for trading BTC and ETH on Binance where decisions come from backtest-calibrated rules and the AI can only veto. Eight agent roles, models chosen by score, and risk and testnet gates before production.
A year maintaining Tucu UI, my open source React component library with macOS-style layout variants. I share the API design, documentation, and versioning decisions behind it, and what I would do differently.
An AI chatbot isn't always the best option for a small business. I share the criteria I use to decide when to automate customer support and when a form or a person does a better job, including recurring costs and risks.
The factors that define the scope and effort of an online store in Argentina: platform or custom build, integrations, electronic invoicing and design, and how I evaluate them before starting a project.
As a Forward Deployed Engineer, I took part, together with a team, in building an investment and trading platform for an energy-sector company with its operations backoffice. I share architecture and process learnings, with no code or internal data.
After measuring tokens per task, the next step is understanding how much each task solved by AI agents costs. I share how to choose the model per task, what to log when a task closes, and what to look at on a cost dashboard.
A more practical second part: how we split domains in an Nx monorepo with microfrontends, what contracts we defined between them, how we versioned them, and what got complicated in CI as it grew. Decisions, trade-offs and takeaways.
The spec template I use daily with AI agents, with the reasoning behind each section: summary, acceptance criteria, and the gates that prevent code from moving forward without a specification. Also how I choose between flow lite and full, and why I log token consumption.
I explain, from my experience as a Forward Deployed Engineer, how this profile differs from a consultant, a product developer, or a solutions engineer. I also share what skills it combines and when it can make sense to add one to a team.
How I migrated a site to bilingual (ES/EN) with app/[lang] routes, proxy rewrites, and reciprocal hreflang without breaking indexed URLs, and how an SDD cycle with plan, tasks, and production verification helped contain the pitfalls along the way.
How I proposed adopting Spec-Driven Development in a YPF development team and how we implemented it together: specs, cycles, role-based agents and gates, which flows we used by change size, and what we learned.
I'll tell you how I went from building chatbots that only answer questions to designing AI agents that operate on a platform's real entities, with function-calling, custom RAG, and a human confirmation step before executing any action. Also, how we control cost by measuring before spending.
An end-to-end walkthrough of DisplayAds, a production product that spans a real-time NestJS API all the way to Raspberry Pi and kiosk-mode Android apps. I share the 8 pieces of the system, how remote updates work with autonomous rollback, the tech stack, and how I kept the project organized with Spec-Driven Development and per-agent cost telemetry.
A first-hand account of Tuki, the RAG-powered quoting assistant running in production at estebanburgos.com.ar/quote. I go over the architecture decisions, why I chose pgvector without an HNSW index, how I keep costs under control with rate limiting, and what broke along the way.
Discover what LangChain is and why it has become one of the central pieces for building modern intelligent chatbots. We go over how to combine RAG, chat-tools, and a smooth user experience to create assistants that truly solve problems, not just answer questions.
The debate on whether artificial intelligence will replace developers is often framed as a false dilemma: **the traditional technical expert versus the average user empowered by an intelligent agent**. However, the real friction is neither syntactic nor superficial. It's not about who writes a Python function faster or who sets up an automation flow in half an hour. The real gap is **architectural and logical**.
The speed at which artificial intelligences can write code today is astonishing, but it comes with a very high hidden cost: **technical anarchy**. When multiple developers integrate AI agents into their daily workflows in an improvised manner, the result is often inconsistent code, silent regressions, undocumented patches, and an alarming loss of traceability. Ultimately, a new type of **"agentic spaghetti code"** that stifles long-term software maintainability. To solve this problem at its root, I designed **SDD-Harness**, a portable, language-agnostic command-line interface (CLI) framework that implements the **Spec-Driven Development (SDD)** methodology. The premise is as simple as it is inflexible: **no line of code is written without a formal, structured, and pre-validated technical specification**. Below, I share a deep dive into how the tool works and why agent governance is the indispensable step to scale AI-assisted software engineering.
Discover how microfrontends are transforming software development in the era of shared, SDD-driven agentic systems. We explore the flexibility of building interfaces as independent apps or as apps co-located with their BFFs in a monorepo, orchestration with Nx, and the efficiency of shared code through libraries: a paradigm well worth admiring.
Discover Spec-Driven Development (SDD), the AI-First methodology that transforms product vision into precise and auditable code. In a world where AI is key, SDD ensures that every line of code is aligned with your business objectives, surpassing the limits of traditional development.
Efficient token management is a crucial challenge in AI development. This article explores how agentic systems can empower agents to define the model, token allocation, and the effort required for a task, while ensuring impeccable quality. Discover how to optimize costs and resources without compromising excellence in your artificial intelligence solutions.
The Software as a Service (SaaS) model has dominated the technological landscape, offering standardized solutions for a wide range of needs. However, its 'one-size-fits-all' approach is yielding ground to the rise of Artificial Intelligence development, which is creating hyper-personalized and truly useful tools. This evolution promises a new era of technological solutions focused on the specific needs of each client.
Artificial intelligence is redefining software development, transforming development hours into 'AI tokens' and creating a dilemma in estimations. Both clients and developers face a new paradigm: how do we value work when the machine generates the code? This article explores how we should approach quotes today, focusing on strategic value and experience.
In the era of artificial intelligence, we need our code to speak the same language as AI. Discover why generating MPCs is fundamental for your personal projects, how AI can become your best ally, and how I have transformed my frontend workflow with Tucu-ui MCP.
I am very excited to share the latest update of tucu-ui. I have designed new layout variants inspired by macOS so that your web applications look clean, modern, and super professional with zero effort.
Gatsby JS is a static site generator used to create high-performance websites. Built on React, it combines some of the best p...