Workshop on Sourcing with AI

Glossary

Assistants, APIs, data vendors, and the words IT uses

Claude

Anthropic's assistants and APIs for chat and automation. In sourcing we use Claude for long-context drafting from notes when policy allows redacted snippets, never as silent autopilot on candidate send.

Microsoft Copilot

Microsoft 365 embedded assistant in Word, Outlook, Teams, and related surfaces. Fits teams where first-touch drafts and edits already happen inside your Microsoft tenant and retention rules.

Codex

OpenAI's coding-oriented assistant used inside editor and agent flows to edit files, run commands, and keep changes in version control. Sourcers meet it when ops or a technical sourcer maintains parsers, n8n steps, or small glue next to templates.

Claude Code

Anthropic's coding agent that works across a repository from the terminal or editor. Same guardrails as other coding agents: strong for playbook files and integration scripts, weak as a substitute for Legal or brand review on outbound.

ChatGPT

OpenAI's chat products, including enterprise workspaces when your company enables them. Common first stop for outreach variants and search rationale. Still needs your written sign-off path before anything candidate-facing ships.

Gemini

Google's assistant and Workspace integrations. Fits drafts and research when mail and docs already live in Google and your admin allows the data classes you paste.

Cursor

An IDE-first assistant for editing files with diffs and project context. Useful when your team treats sourcing playbooks, rubrics, and small scripts as versioned text instead of a slide deck.

API

A published contract so one system asks another for data or actions over HTTPS, usually with keys scoped to least privilege. We label which calls are read-only, which write to ATS, and which must never carry raw candidate exports.

Integration

Purposeful wiring between approved systems: ATS, CRM, scheduling, enrichment, or a hosted model endpoint. Different from exporting CSVs into a personal chat tab that IT cannot see.

Webhook

An HTTPS callback one system sends another when an event happens, for example a new tag in ATS or a form submission. Common glue when you move from copy-paste to light automation with explicit owners.

Data providers (examples: Coresignal, Bright Data)

Vendors that sell datasets or feeds of company and people signals for research and list building. Useful for market context when contracts and purpose limits are clear. Still not permission to ignore consent, employer rules, or minimization in prompts.

Data enrichment (examples: ContactOut, Findymail)

Vendors that append emails, phones, or other fields to profiles you already found. Fast for routing when accuracy is high; expensive and policy-heavy when every prospect is auto-enriched without a human spot check.

LLM (large language model)

The model behind chat assistants: it predicts the next tokens from your prompt and context. It does not know your private ATS state unless you connect it through an approved integration or paste policy-safe snippets.

n8n

Open-source style workflow automation (often self-hosted or cloud) that chains HTTP requests, transforms, and schedules. Shows up when teams graduate from manual copy-paste into owned automation with logs and keys.

Human-in-the-loop

Named reviewers before external send, before writes to ATS, or before candidate identifiers leave approved surfaces. We write these as checklist steps with owners, not as a buzzword on a slide.

PII minimization

Share only the fields a task needs, redact or pseudonymize in classrooms, and keep bank-grade or health data in systems built for them. Applies twice when models and enrichment vendors both touch the same person record.

Model drift

Model behavior shifts when providers ship new versions. Mitigation: pinned prompts, dated playbook reviews, and paired checks on search drafts plus outreach after each major upgrade.