Ideas from
the work.

Experience-based perspectives on AI, technology decisions, infrastructure, and the operational work behind lasting change.

FIG. 01

AI & Automation Oct 22, 2026 10 min read

Agent memory in practice: the four layers we run, and why we refused a fifth

AI agent memory is not one system. We run four, each solving a different forgetting problem, and each with its own way of failing quietly. What each layer is for, how it can fail and the rule that keeps it honest.

FIG. 02

AI & Automation Oct 22, 2026 9 min read

A RAG chatbot on your own content: the decisions that matter after the demo

A RAG chatbot answers from your pages and documents instead of from the model's memory. Building one takes an afternoon. These are the product and operations decisions we made running ChatWise in front of real customers.

FIG. 03

AI & Automation Oct 22, 2026 11 min read

The LLM Wiki in production: four months on Karpathy's pattern, and the rules that keep it working

We run our company on an LLM Wiki: an Obsidian vault that Claude Code and other agents maintain alongside two humans. More than 660 commits in, here is what carries the value, the rules that keep it working and what to set up from day one.

FIG. 04

AI & Automation Oct 22, 2026 8 min read

What is an AI agent, and when does your business need one?

An AI agent is a language model with tools and a loop that decides the next step toward a goal. How it differs from a chatbot and from an automation, the agents we run in production, and a checklist for deciding whether your business needs one.

FIG. 05

AI & Automation Oct 22, 2026 9 min read

What an AI operating system is, for a business: the layer we run our agents on

An AI operating system for a business is the shared layer that lets a company run many AI agents safely: memory, one front door for work, procedures, scopes, approvals and monitoring. Here is each part, what breaks without it and how to start.

FIG. 06

AI & Automation Oct 22, 2026 8 min read

What is RAG? Retrieval augmented generation explained with a real pipeline

RAG is the technique that lets a language model answer from your content instead of from what it remembers from training. Every step, from indexing to answering, traced through the real code of ChatWise.

FIG. 07

AI & Automation Oct 22, 2026 12 min read

WhatsApp AI agents in production: eight failure modes and the fixes that work

AI agents in WhatsApp groups, support inboxes and team channels tend to fail in ways the prompt says they cannot. These are eight failure modes, why each one happens and the code-level fix for each.

FIG. 08

Paying a lot, getting little Jul 19, 2026

Hype is making your AI decisions (and it decides badly)

A big share of bad AI investments don't come from bad technology, but from decisions built on wrong ideas: inflated expectations, fear of falling behind, and demos that aren't products. The misconceptions we see most, and how to decide better.

FIG. 09

Digital Transformation Jul 19, 2026

The inverted revolution: AI adoption is happening from the bottom up

Every previous technology wave entered companies from the top: leadership bought it, then pushed people to use it. AI came in the other way around: your team already uses it. That completely changes what managing change means.

FIG. 10

Paying a lot, getting little Jul 9, 2026

"I pay a ton of money and things still don't work"

It is the sentence we hear most often in first meetings with clients. It is almost never a budget problem: it is a problem with how the relationship with your technology vendors is set up. Here is what we see, and what you can do about it.

FIG. 11 $What is a fractional CTO, and when do you need one?$

Paying a lot, getting little Jul 9, 2026

What is a fractional CTO, and when do you need one?

A fractional CTO is senior technical judgment on your side of the table, for a fraction of the time and cost of a full-time hire. What they do, the signals that you need one, and when it is not the answer.

FIG. 12

Software Development Apr 12, 2026 6 min read

Why we stopped mocking databases in integration tests

We burned an afternoon chasing a migration bug that mocked tests had cheerfully approved. Here's the policy we shipped instead, and the cost-benefit it implies.

FIG. 13

AI & Automation Mar 28, 2026 9 min read

The evals you actually need before shipping an LLM feature

Most teams either over-engineer this (full eval framework, dashboards, three contractors) or under-engineer it (vibes). Here's the minimum we ship with.

FIG. 14

Cloud & DevOps Feb 18, 2026 11 min read

Akamai vs Cloudflare for LATAM traffic: a 90-day side-by-side

We ran the same Costa Rica e-commerce site on Akamai (via Linode) and Cloudflare for three months. Here's where each one earned its keep, and where it didn't.