What’s the difference between an AI agent and an automated workflow?
An AI agent interprets information and can select actions within defined rules and permissions. An automated workflow follows a predetermined sequence of steps. AI agents are useful when a process involves variable information or decisions that require interpretation.
For predictable tasks, a simpler automation can be more appropriate. SLIDEFACTORY develops both, using AI workflow automation where the process requires it.
How long does it take to develop an AI agent?
The schedule depends on the workflow, integrations, data access, testing requirements, and complexity of the system.
For larger or uncertain projects, we may recommend a scoped pilot with agreed success criteria before moving into broader development. The implementation schedule is established during planning.
What does an AI agent cost to run?
Running costs depend on the models used, the amount of information processed, how frequently the agent operates, and the supporting infrastructure.
For reference, our CodeRaven code review agent used $161.20 in model processing costs across 2,812 review runs over approximately five months. That figure covers model processing for that system, rather than a general estimate for other agents. We track usage and compute costs in production.
How much does AI agent development cost?
SLIDEFACTORY’s AI Opportunity Assessment starts at $1,500 and includes discovery, workflow and systems review, prioritized recommendations, and an implementation roadmap.
Agent development is scoped and quoted after the assessment. The cost depends on the workflow, software integrations, security requirements, and development involved.
Will AI agents replace our employees?
Our approach focuses on the repetitive work that sits between receiving information and making decisions. That can include sorting requests, retrieving documents, moving data, and preparing information for review.
People remain responsible for decisions that require their judgment. We design human approval checkpoints around the risk of each action.
Can you take over an AI agent another vendor started?
Yes. SLIDEFACTORY can review existing AI agents, stalled implementations, and AI-generated codebases. We begin with an audit to understand the system, identify problems, and determine which parts are worth keeping.
We can then recommend how to improve, complete, or integrate the existing work.
Can we use AI automation without building a full agent?
Yes. Many business processes can be improved with standard software integrations, predefined workflows, and selective use of AI.
If you came looking for an AI automation agency, our AI automation services cover both simpler workflows and more complex agents. We recommend the approach that fits the process rather than adding complexity where it isn’t needed.
Which AI models do you use?
We work with Claude, OpenAI, and Gemini models, choosing technology based on the needs of each workload.
Our architecture supports movement between providers. We’ve moved production workloads between models, with engineering and evaluation to account for differences in behavior, accuracy, and cost.
Do you build custom LLM and RAG applications?
Yes. Most custom LLM work starts with retrieval-augmented generation (RAG): the system searches your documents, tickets or knowledge base and passes the model only the passages it needs, so answers can point to their source.
We build these as chatbots, internal assistants or steps inside a larger agent, on Claude, OpenAI or Gemini models. The same approach covers language tasks such as classifying requests, extracting details from documents and summarizing long text.
Can AI agents connect to our existing business software?
Yes. We use standard APIs and the Model Context Protocol (MCP) to connect agents with CRMs, ticketing systems, analytics tools, content platforms, and codebases.
We’ve built MCP integrations for several products and also work with integrations published by platform providers. Where a system has no API, we identify that constraint and its implications before you commit.
How do you manage security and control?
We design agent access around least-privilege permissions and defined data boundaries, with audit trails and human review checkpoints appropriate to the task. Rules governing agent behavior are enforced technically, and a security review is completed before launch.
SLIDEFACTORY has worked in healthcare and wellness since 2018, including an early VR project involving end-of-life management. Security requirements are evaluated for each project and the information it handles.
