An AI agent pursues a goal, uses tools, gathers information, makes intermediate decisions, and attempts to complete work. The business opportunity is not only the reasoning model. It includes the workflow, permissions, integrations, guardrails, monitoring, and human approval system around the agent.
Agent map
| Agent type | Typical work | Opportunity spaces |
|---|---|---|
| Research agent | Find, compare, synthesize, and document information | Industry research, due diligence, procurement, policy, medicine, science, and competitive intelligence |
| Computer-use agent | Operate websites and desktop software through the interface | Legacy-system automation, data entry, claims, government forms, scheduling, and back-office work |
| Coding agent | Plan, write, test, review, and maintain software | Vertical development agents, testing, security, migration, documentation, and non-coder app building |
| Customer-service agent | Answer questions and resolve common requests | Voice, multilingual support, refunds, troubleshooting, scheduling, and account actions |
| Sales agent | Research prospects, prepare outreach, update records, and follow up | Niche markets, local sales, channel partnerships, and compliance-aware outreach |
| Operations agent | Coordinate tasks across systems and people | Logistics, property management, construction, healthcare administration, and field service |
| Finance agent | Reconcile, classify, monitor, forecast, and prepare reports | Bookkeeping, accounts receivable, audit preparation, fraud, and small-business finance |
| HR agent | Support recruiting, onboarding, benefits, training, and employee service | Skills matching, compliance, scheduling, and workforce mobility |
| Security agent | Investigate alerts, gather evidence, propose remediation, and monitor systems | Managed security, small-business defense, identity, agent security, and audit trails |
| Personal agent | Coordinate a user's information, apps, plans, and purchases | Privacy-first memory, family coordination, elder care, travel, and consumer advocacy |
| Scientific agent | Connect literature, code, simulation, and experiments | Biology, chemistry, materials, engineering, and automated laboratories |
| Multi-agent system | Divide work among specialized agents and combine results | Complex projects, verification, parallel search, enterprise operations, and software development |
Representative agent platforms
Salesforce Agentforce
A platform for customer, employee, sales, and service agents grounded in enterprise data.
Microsoft Copilot Studio
A low-code environment for building, deploying, governing, and connecting organizational agents.
Gemini Enterprise
Google Cloud's enterprise search, assistant, and agentic workflow platform.
Amazon Bedrock
AWS infrastructure for building and operating generative AI applications and agents.
GitHub Copilot Agents
Coding agents that can take assigned development work and prepare changes for review.
OpenAI Codex
A software-engineering agent for code changes, reviews, migrations, automation, and multi-agent development.
Where new companies can still win
- Vertical specialization: one industry's terminology, rules, data, and systems
- Reliable completion: charging for resolved work rather than generated text
- Agent identity and permissions: controlling exactly what an agent may see and do
- Evaluation: realistic testing, simulation, failure analysis, and performance measurement
- Human approval systems: routing risky actions to the right person with clear evidence
- Memory and provenance: remembering useful context while showing where information came from
- Integration: connecting agents to old software, phones, documents, and physical operations
- Agent security: preventing prompt injection, credential misuse, data leakage, and unauthorized action
- Agent marketplaces: discovery, reputation, certification, billing, and interoperability
- Small-business agents: affordable systems that do not require enterprise consultants
Important distinction
An agent demonstration proves that a task can sometimes be completed. A business requires the task to be completed reliably, economically, securely, and with accountable supervision.