Agents we have built
Executive assistant for private equity: answers from the ERP, email and files with sources, and never changes anything without approval.
Multi-agent e-commerce campaigns: a manager agent splits each request across agents for analysis, promo codes, copy and design, so a non-technical team launches a campaign from one chat.
Legal research assistant: searches court decisions and legislation and answers with citations to the decisions it found.
Upsight: our own product, which researches Upwork jobs and drafts proposals overnight for agencies.
What makes an agent safe to use
A narrow job: one workflow with a clear start and end, not a general assistant.
Grounded answers: the agent reads your documents and systems and cites where each answer came from.
Limited permissions: read access by default; anything that writes, sends or pays goes through a person.
Logs and evaluation: every run is recorded, and we test the agent against real examples before and after launch.
Agent, chatbot or automation
A fixed process with predictable steps is usually better as a plain automation: cheaper, faster and easier to debug. A chatbot answers questions. An agent decides which steps to take, so it earns its cost when the path changes from case to case. Our article What is an AI agent, really? explains the difference, and What does an AI agent cost to run? covers the running costs.
