You shipped a chat widget. It works. The team trusts it. Most inbound questions get answered with the source attached and your team approves anything that crosses the boundary into a CRM record or a customer email. Now the question on the next planning call is: what about agents?
The good news is that the answer is not “build a second platform.” The path from chat to agents on Atlas is the same substrate doing more work. The knowledge base you spent the last year curating is the same one the agents read from. The approval seat is the same one your operator already uses. The audit log captures both. The economics shift (the chat product is conversation-priced; the agent product is per-active-agent-priced), but the architecture does not.
The progression Legacy ran
The cleanest worked example of this progression is Legacy, a healthcare growth team that started on Atlas at go-live 12 months ago.
Month one. Email-only support deflection.Atlas drafted responses to inbound customer email; Legacy’s team approved each one. The smallest possible thing. It built the muscle: what does Atlas do well, what does it get wrong, what do the source documents actually say versus what the team thought they said.
Months two to four. Web chat.The next channel. Same knowledge base, same source receipts, same approval boundary, different surface. 6.1x more inbound conversations on the same traffic, 25.2% buyer-intent rate, ~94% deflection on repetitive questions. The substrate compounded; the operator’s daily work expanded but did not become a different skill.
Months five to nine. Voice agents. Inbound phone calls handled in the moment with the same knowledge base and the same approval boundaries. Voice is harder than chat because the channel is real-time; the operator work shifted from approving each draft to flagging anything the agent should not repeat. Same substrate, new review pattern.
Months ten to twelve. Integrated agents. Agents that read from the CRM and write back to it. Agents that talk to the kit-ordering system. Both built from connectors in the Clarm catalogue, not from custom engineering. This is the step most teams stall on, because integrated agents are the ones that can break things. The reason it worked here is the previous nine months. By the time the integrated agents shipped, Legacy already trusted the approval queue, already trusted the source receipts, already had a year of audit-log evidence about how the system behaves.
Twelve months in, total case volume across email, chat, voice, and integrated workflows is roughly 8x what it was at go-live. Legacy’s team has been in the approval seat the entire 12 months.
What you are buying when you move from chat to agents
Three things change technically. One thing changes commercially. Everything else stays the same.
Triggers. Chat runs in response to a visitor message. Agents run on a schedule or in response to an event (a customer email lands, a CRM record changes, a calendar event arrives, a webhook fires). The trigger system is part of the Atlas substrate, exposed to operators as templates rather than code.
Connectors.Chat reads. Agents read and write. Atlas’s connector catalogue covers Microsoft Graph, Salesforce, SAP, Avaloq, Aladdin, Slack, Teams, plus the standard HTTP / REST / generic-DB patterns. When you add a connector, both the chat surface and the agent surface can use it.
Approval-queue timing. The chat operator reviews answers after the visitor saw them, or marks specific topics for pre-approval. The agent operator reviews drafts before any external action happens. Same seat, different timing.
Commercial. The chat product runs on the Free or Growth tier, priced by conversation. Agent workflows run on the Atlas / Enterprise tier (per-customer pilot then monthly subscription, with the pilot fee crediting against the first months of subscription). The boundary is honest because the work is different; the substrate that powers it is the same.
What does not change
- The knowledge base. Same documents, same source receipts, same retrieval, same access controls.
- The compliance posture. SOC 2 Type II, HIPAA, Swiss FADP, GDPR. The substrate-level controls apply identically to chat and agent surfaces.
- The bring-your-own-LLM choice. Whichever model you swapped in for the chat product is the model the agents use, until you swap again.
- The audit log. One log, two consumption modes. Your auditor sees a unified view of every agent action and every chat answer.
- The operator’s seat. The same person, with the same approval authority, doing the same kind of work.
What to ship first
The most common first agent after a successful chat year is one of three:
- Inbound email deflection. Reuses the chat knowledge base directly. Operator approves drafts before send. Smallest possible second workflow.
- Voice-to-CRM. Operator finishes a customer call, dictates a 60-90 second voice memo, Atlas drafts the CRM update, follow-up email, and internal chase. Operator approves. Most common starting workflow for client-facing teams.
- Scheduled briefing pack. Atlas reads from the systems the team already uses (calendar, CRM, support history) and assembles a morning briefing for review. Operator scans, approves, and the briefing lands in Slack or Teams.
The shape of all three: the operator’s daily work becomes faster, not different. The agent drafts; the operator approves; the substrate captures everything.
The honest version of the upgrade conversation
If you have been on the chat product for a year and are considering the agent move, the honest version of the conversation is short. You already have the substrate. You already have the operator muscle. You already have the audit log your compliance team trusts. The agent move is your team deciding which workflow goes second.
The technical migration is small. The commercial step is moving to the Atlas / Enterprise tier (per-customer pilot, then monthly subscription that includes the substrate platform fee and per-active-agent opex). The pilot fee credits against the first months of subscription, which makes the commercial step a clean one-month decision rather than a multi-quarter procurement cycle.
For the architecture, read What Is Atlas?. For the case study, read How Legacy Went 8x in 12 Months on Atlas. For the buying-side checklist, read The Agent-Deployment Buying Guide. When you are ready, book a pilot discussion.