Glossary of software development with coding agents
This glossary defines, in one or two sentences each, the 20 terms Ithaca uses to talk about software development with coding agents, from coding agent and harness engineering to level 4, dark factory and session record. Each term links to the page that covers it in depth; four of those pages are in Portuguese and are marked.
Published
Agentic development platform
The system in which coding agents write most of the code while the team writes the specification, sets the limits of the agents' autonomy and decides what reaches production. Every session is logged, with its authorship.
Agentic development platform: what it is
Agentic engineering
A way of building software in which coding agents write most of the code and people orchestrate and oversee them: they specify, set limits, check and approve. In February 2026, Andrej Karpathy called "agentic engineering" his current favorite name for the professional way of programming through LLM agents, the one with "more oversight and scrutiny."
Agentic development platform: what it is
AI governance in software development
The set of rules that defines what coding agents may do without prior approval, what needs a person's sign-off first, and who answers for the code that goes to production.
AI governance in software development
Autonomy levels
A scale from 0 to 5, proposed by Dan Shapiro in January 2026, that describes how much of the work of programming moves from a person to AI. It runs from level 0, where nothing is written without the programmer's approval, to level 5, the dark factory.
Autonomy limit
What a coding agent may do without prior approval and what needs human approval before it runs, set by the team by type of system and by criticality. Reading what goes to production applies to everything.
AI governance in software development
Characterization test
A test that documents what a system does today, not what the documentation says it should do. Michael C. Feathers described the practice in Working Effectively with Legacy Code, and it anchors the check before any part of a legacy system is replaced.
Refactoring with AI without losing accuracy (in Portuguese)
Coding agent
An AI agent that takes over parts of the execution: it reads the repository, proposes changes, opens pull requests and runs tests. Unlike a coding assistant in the IDE, which speeds up the person already typing, it needs an autonomy limit and a session record.
AI governance in software development
Coding agent observability
The structured record of every session in which an agent works on a company's code: the request, the steps it took, the model it used, the cost, the result and the authorship. With that record, a company can audit how AI is used in development and see what to improve.
Observability for coding agents
Dark factory
On the factory floor it is the plant that runs with the lights out, because nobody is there to see anything. In software, it is the code nobody reads before it goes into production. On Dan Shapiro's scale, it is level 5.
Harness engineering
The work of building the environment an AI agent operates in: the context it receives, the tools it can call, the rules that apply to it, and the point where its autonomy stops.
Harness engineering for enterprise coding agents
Human in the loop
On an agentic development platform, people decide at the points that carry weight: the specification, what the agents may do without prior approval, and what goes to production. On Ithaca's platform, that is what keeps the work at level 4 and out of the dark factory.
Agentic development platform: what it is
Knowledge base
The first stage of the platform: the codebase, documentation and anonymized data go in before any new line is written. The context the team approves feeds the next session.
Agentic development platform: what it is
Legacy modernization
With AI, it means using coding agents to read what a system does today and build its replacement alongside it, switching over piece by piece while the old system keeps running.
Legacy modernization without stopping the business
Legacy system
Software that still runs an important part of the business but that the team can no longer change safely and at a reasonable cost, for lack of tests, current documentation, people who know the code, or support for the technology. Its age matters less than the cost of changing it.
Legacy systems: what they are (in Portuguese)
Level 4
On Dan Shapiro's scale, the level where the person writes and argues over the spec and comes back later to see if the tests pass. Ithaca's platform operates at this level with one more rule: people on the team read what goes to production, and every session is logged.
Refactoring with AI
Using coding agents to rewrite parts of a system without changing the behavior the business still depends on. On legacy systems, accuracy comes before speed: characterization tests, human reading of what goes in, and a session record.
Refactoring with AI without losing accuracy (in Portuguese)
Session record
The record each coding-agent session leaves: request, steps, model, cost, result and authorship. Months later, it still shows who decided what.
Observability for coding agents
Software development with AI
At Ithaca, it is work done by coding agents within the limits the client's team sets. People on the team read the code before it goes to production, and every session is logged.
Ithaca: coding agents with human oversight
Strangler fig
A modernization path in which the new system is built beside the old one and takes over its functions gradually, until the old one can be switched off. Martin Fowler named it after the strangler fig, a vine that grows around a host tree and can end up replacing it.
Legacy systems: modernization paths (in Portuguese)
Sustainment
The stage of the platform after go-live: tickets come in and fixes go out, with the team supervising. It is part of the platform from the start.
Agentic development platform: what it is
Frequently asked questions
What is the difference between a coding agent and a coding assistant?
A coding assistant speeds up the person already typing in the editor. A coding agent takes over parts of the execution: it reads the repository, proposes changes, opens pull requests and runs tests. That is why an agent needs an autonomy limit and a session record.
How is harness engineering different from observability?
The harness prepares and limits the execution: context, rules, tools and the point where autonomy stops. Observability records what happened in the session: request, steps, model, cost, result and authorship. Neither one replaces the other.
Are a dark factory and level 5 the same thing?
Yes. On Dan Shapiro's scale, dark factory is the name for level 5, where the process is "a black box that turns specs into software" and nobody reads the code. Ithaca's platform runs at level 4 and stops before it.
What goes into a session record?
Six fields: the request and who made it, the steps the agent took, the model used at each step, the cost, the result (a diff, a pull request or a rejection) and the authorship, meaning who asked, who executed and who supervised.
