$ cat ./glossary/definitions.md

The AI agent glossary

Crisp, citable definitions of the terms that come up when you build with agents. Short enough to quote, precise enough to use.

$ grep -R "^## " ./glossary

Terms, defined plainly

definitions.md
Agentic AI
AI systems that plan, use tools, and take multi-step actions to complete tasks with limited human supervision.
AI agent
Software that perceives its environment, reasons, and acts toward a goal with a degree of autonomy.
ASDLC
The Agentic Software Development Lifecycle: BuildHaus's structured method for specifying, architecting, building, verifying, and shipping agent systems. read more
Context window
The amount of text a model can consider at once, measured in tokens.
Embeddings
Numeric vectors that represent the meaning of text, so similar content can be matched by similarity.
Evals
Evaluation tests that measure how reliably an agent completes a task before it ships.
Fine-tuning
Adapting a pre-trained model to a specific task by training it further on targeted data.
Function calling
An agent's ability to invoke external APIs and functions, so it can act rather than only generate text.
Grounding
Anchoring a model's output in retrieved, verifiable data rather than its training memory.
Guardrails
Controls that keep an agent's behaviour within safe, compliant, and predictable bounds.
Hallucination
A fluent, confident model output that is factually wrong or unsupported.
Human-in-the-loop
A design where a human approves or steers an agent's highest-stakes decisions.
LLM
Large language model: a model trained on text at scale to predict and generate language.
Model-agnostic
Built to work with any underlying model provider, with no lock-in to a single vendor.
Multi-agent system
Multiple agents that divide work, coordinate, and hand off to complete a task together.
Observability
The ability to trace and inspect what an agent did and why, so failures are diagnosable.
Orchestration
Coordinating the flow of work across tools, models, and agents to complete a task.
Prompt
The instruction and context given to a model that define the task it must perform.
RAG
Retrieval-augmented generation: grounding a model's answers in retrieved documents to reduce hallucination.
Token
The basic unit a language model reads: roughly a word or part of a word.
Vibe coding
Building software by prompting AI without engineering rigour, tests, or review.
Workbench
BuildHaus's open-source framework for agentic software development: project structure, agent workflows, and quality gates. read more
$ man agent-questions

Asked often

What is an AI agent?

An AI agent is software that perceives its environment, reasons, and acts toward a goal with a degree of autonomy. It plans, calls tools and APIs, and carries out multi-step work toward an outcome, rather than only answering a single prompt.

How is an AI agent different from a chatbot?

A chatbot answers one message at a time. An agent plans a task, calls tools, and completes multi-step work toward an outcome. The short version: a chatbot responds, an agent acts.

How much does it cost to build an AI agent?

It depends on scope and risk. A validated prototype can cost a few thousand dollars; a production agent system with guardrails, evals, and handover typically runs from tens to hundreds of thousands. The better question is the cost of getting it wrong. Book a discovery call and we will scope it honestly.

How long does it take to build a production AI agent?

A scoped prototype takes weeks; a load-bearing production system takes months. BuildHaus works to the ASDLC so the timeline is predictable: specify, architect, build, verify, ship.

What is the ASDLC?

The Agentic Software Development Lifecycle: a structured method for shipping agent systems in five stages - specify, architect, build, verify, ship. It applies engineering discipline to agentic work so impressive demos do not become unmaintainable liabilities.

What is the difference between agentic AI and RPA?

RPA automates rigid, rule-based steps. Agentic AI handles variability: it reasons, adapts, and uses tools where the path is not fully scripted. RPA clicks buttons; agents make decisions.

$ book --type discovery --duration 30m --cost $0

Thirty minutes.
Zero dollars. Real answers.

Bring the problem, the half-built prototype, or just the hunch that agents could change how you work. We’ll tell you straight what’s worth building — and what isn’t.

Book a free 30-min discovery call->

# no deck, no pressure, no obligation. exit code 0 either way.