Blog
Software is a lifecycle, not a prompt
Field notes on modernizing legacy systems, governing AI, and taking prototypes to production.
-
Moving the data is the scary part. Here is how we make it dull.
Rebuilding an old system gets the attention, but moving its records is where modernization projects get hurt. Why legacy data migrations fail weeks after cutover, and the rehearse-and-reconcile process Sapilon uses to catch those failures while nobody depends on the answer yet.
-
Design, Build, Rehearsal, Live: why the fastest AI software ships in stages
AI can produce an application in an afternoon, which makes it tempting to deploy it the same afternoon. The four stages of a Sapilon project exist because generation speed was never the bottleneck. Verification is, and stages are how verification keeps up.
-
How much does a custom internal tool cost?
Agencies quote $30,000–$150,000. Low-code quotes $30 per seat per month, forever. We break down what an internal tool really costs across building, running, and changing it, and what the same tool costs on an AI-first platform.
-
AI-first doesn't mean AI-only: why Sapilon has human experts
Most of a software project is work an AI agent handles well. A small number of moments are not: new dependencies, schema changes, infrastructure, going live. Sapilon Experts exist for exactly those moments, and for nothing else.
-
AI-safe code: ownership zones explained
Sapilon codebases answer "what is the AI allowed to change?" in the code itself. Three tags (//!AUTO, //!SAFE, and //!DEV) mark who owns each part of the system, and the tooling enforces the boundary on every change.
-
Your old system already wrote the spec: how Sapilon thinks about artifacts
Every modernization project starts with a folder: a database dump, screenshots of the old UI, an Excel export, a Word doc from 2014. Most tools treat those files as clutter. We treat them as the requirements document nobody ever wrote.
-
Is this the end of low-code and no-code?
Low-code promised software without engineers, and delivered speed at the price of ceilings and lock-in. AI-first systems now offer the same accessibility on top of real code. What survives, and what doesn't?
-
Why prompts are not enough for creating a production app
Prompt-to-app tools produce impressive demos in minutes. But production software is a lifecycle: security, data, operations, and change over years. A prompt has no opinion about any of it.
-
AI agents now write software. Who's flying the plane?
Coding agents went from autocomplete to opening pull requests on their own. Aviation solved this problem decades ago, not by removing the pilot but by defining exactly what autopilot may do. Your codebase needs the same answer.