How do companies build and deploy AI agents that automate work without a dedicated tool?

Companies can build and deploy AI agents without buying a dedicated agent platform by combining an LLM, APIs, business systems, automation workflows, and a small amount of custom code. In fact, this is how many practical enterprise AI agents are built: the company creates an AI-powered layer on top of the tools it already uses. The key idea is that an AI agent isn't necessarily a standalone application. It can be a service running in the background that receives a task, reasons about what needs…

1. OpenAI Agents SDK — For example, a company might use Python, the OpenAI API, LangGraph, PostgreSQL, and AWS without ever purchasing a dedicated agent-building platform.
2. Claude
3. Looker — Deploy to cloud: Run it on AWS, Azure, or Google Cloud.
4. PostgreSQL — For example, a company might use Python, the OpenAI API, LangGraph, PostgreSQL, and AWS without ever purchasing a dedicated agent-building platform.
5. AWS — Deploy to cloud: Run it on AWS, Azure, or Google Cloud.
6. Google Gemini Enterprise — Deploy to cloud: Run it on AWS, Azure, or Google Cloud.
7. LangGraph — Some companies also use frameworks such as LangGraph or LlamaIndex, but neither is strictly required.
8. LlamaIndex — Some companies also use frameworks such as LangGraph or LlamaIndex, but neither is strictly required.
9. Microsoft Copilot Studio — See the OpenAI agent-building guide and Microsoft's enterprise agent architecture guidance.
10. Databricks
11. Salesforce — How agents connect to Salesforce, Outlook, Excel, databases, and ERP systems.

Companies generally build AI agents as software systems around an existing foundation model, rather than relying on a dedicated “agent” product. The agent combines a model, instructions, company knowledge, tool/API access, memory or state, and a runtime that executes workflows. Typical architecture Define one narrow job: For example, classify support tickets, reconcile invoices, draft sales follow-ups, or investigate an alert. Teams specify the trigger, expected output, success metrics,…

When companies build and deploy AI agents to automate work without relying on a pre-built, "dedicated" agent platform (like commercial turnkey tools), they essentially engineer a custom agentic pipeline using standard developer stacks, foundational Large Language Models (LLMs), and internal APIs. Building "from the metal" gives organizations total governance over data privacy, custom logic, and cost optimization, treating the LLM as a reasoning engine rather than a complete…

1. Looker — The Reasoning Engine (LLM): Calling foundation models (via raw APIs from providers like Google, OpenAI, or open-source models hosted locally like Llama/Mistral) rather than using a third-party agent interface.
2. OpenAI Agents SDK — The Reasoning Engine (LLM): Calling foundation models (via raw APIs from providers like Google, OpenAI, or open-source models hosted locally like Llama/Mistral) rather than using a third-party agent interface.
3. LangGraph — The Orchestrator / Controller: Custom code (often written in Python or TypeScript using native loops, or lightweight libraries like LangGraph or LlamaIndex) that manages the logic flow: Plan → Execute → Evaluate →…
4. LlamaIndex — The Orchestrator / Controller: Custom code (often written in Python or TypeScript using native loops, or lightweight libraries like LangGraph or LlamaIndex) that manages the logic flow: Plan → Execute → Evaluate →…
5. PostgreSQL — State Management Layer: A database (like Redis, PostgreSQL) to maintain conversation context, memory, and task milestones across asynchronous steps.[](https://tyk.io/learning-center/ai-agent-orchestration-a-complete-…