How do you build sales agents with AI agents?

Key facts

Pattern	Research → summarise → draft → human approval → log
Runtime	Function calling loop on /v1/chat/completions (live)
Tools	CRM lookup, news and filing search, email drafting, task creation
Models	30+ models behind one endpoint, route research and drafting separately
Controls	Scoped keys per integration with RBAC and SSO/SCIM
Approvals	No message sends without explicit rep approval
Audit	Per-turn tool and model traces for every draft
Free tier	plugsky-micro and plugsky-lite on the free plan, no card required

TL;DR

• Automate research, summarising and CRM hygiene first — those are safe wins.
• Draft outreach for reps to edit; never send autonomously.
• Enrich accounts through tools so facts come from systems, not model memory.
• Keep data minimisation strict: contact details only where the task needs them.
• Measure time saved and reply quality, not messages sent.

How it works, step by step

1. Pick a bounded workflow — pre-meeting briefs, account summaries or CRM cleanup — and define what the agent may touch.
2. Expose CRM reads, filing search and task creation as tools with typed schemas and least-privilege scopes.
3. Implement the loop: send the request with a tools array, execute tool_calls, append results, and stop at a draft.
4. Require structured drafts — subject, body, personalisation facts used, suggested next step — via JSON mode.
5. Route every draft to a rep for edit and approval; keep sending outside the agent entirely.
6. Log tool calls, drafts and edits so quality and compliance can be reviewed.
7. Evaluate against a baseline of rep-written work on time saved, edit distance and reply outcomes.

Try it yourself

Open the AI agent builder →

Tools for research, drafting and CRM hygiene

The highest-value sales agent is not a writer; it is a researcher and a record-keeper. It gathers account context, summarises the last twelve months of activity, identifies open commitments and drafts a brief before a call. Those tasks are read-heavy, easy to verify and immediately useful.

Plugsky's live function calling supports the loop: a tools array on /v1/chat/completions, tool_calls executed against your CRM and content systems, results appended, iteration until the draft is ready. Because the API is OpenAI-compatible, the tools reuse existing integrations and credentials.

Human approval and compliance

Outreach is a regulated, reputation-bearing activity. Keep the agent on the drafting side: it proposes, a rep approves. That boundary also keeps you clear of automated decisioning concerns and preserves the personal judgement that makes outreach effective.

• No autonomous sends: the send button belongs to a person, always.
• Data minimisation: include only the contact facts the message needs, and keep sensitive fields out of prompts.
• Suppression rules: check opt-outs and do-not-contact lists in tools before drafting.
• Audit: store drafts, the facts used and the approving rep for every message.

Models, metrics and rollout

Route by task: small models for classification, summarisation and CRM field suggestions; stronger models for account synthesis and nuanced drafting. With 30+ models behind one endpoint, the routing table is configuration rather than integration work.

Measure what matters: rep time saved, edit distance between draft and sent message, reply and meeting rates, and CRM completeness. Edit distance is the honest quality signal — if reps rewrite everything, the agent is not helping. Roll out to one team with a clear opt-in, review traces weekly, and expand only when the numbers hold. For customer data restrictions, pin the workspace to a region-locked plane or deploy VPC, on-prem or air-gapped with the same agent code.

Honest comparison

Concern	Plugsky sales agent	Manual research	Email automation tools
Account research	Gathered through live tools	Rep time	Static templates
Drafting	Grounded drafts with facts cited	Rep writes	Merge-field mail
CRM updates	Structured suggestions via JSON mode	Manual entry	Limited
Approval	Rep edits and sends	Rep owns everything	Often automated sends
Audit	Traces of tools, drafts and approvals	Activity history	Send logs

Frequently asked questions

Can the agent send emails?

No, and it should not. Draft for rep approval and keep sending in your existing tools. This preserves compliance and keeps a person accountable for customer communication.

What should the first workflow be?

Pre-meeting briefs or account summaries. They are read-only, verifiable and save time immediately, which builds trust before you automate drafting.

How do we keep outreach accurate?

Enrich through tools so facts come from CRM and approved content, require drafts to reference those facts, and let reps catch anything that slips through.

What data goes into prompts?

Only what the task needs. Avoid sensitive fields, apply suppression checks before drafting, and align retention settings with your privacy policy.

Is function calling live?

Yes. OpenAI-style function calling and streaming are live on chat completions, so the research and drafting loop runs today.

How do we measure value?

Track rep time saved, edit distance on drafts, reply and meeting rates, and CRM completeness. Edit distance tells you whether drafts are genuinely useful.

Can it run inside our environment?

Yes. Plugsky supports region-locked cloud planes plus VPC, on-prem and air-gapped deployment, so customer data can stay within your network.