agenticX — Working Development Example
Human First. Then Agents.
The June 2026 Demo Plan for AgileX Recruitment Intelligence
AgileX / 2nth.ai
June 2026
Google Workspace + Huly ATS + MCP
1. The Principle
"We don't start with AI. We start with your digital workflow. Then we show you where agents fit, what models to use, and how to deploy safely and compliantly. The framework and model choices from
know.2nth.ai are part of the project with you — not a pre-baked decision."
— AgileX positioning to clients and partners
For AI-driven workflows to work in recruitment, there must first be a human-observable digital process that AI agents can participate in. You can't automate what isn't digital. You can't verify AI work without a human-readable trail.
agenticX is the working example: a real recruitment agency dev environment where a human recruiter takes a candidate through the full value chain, and at each step we show where AI agents and tooling can assist, accelerate, or automate.
What the demo proves
- A real candidate moves through a real pipeline
- Every step has a digital artifact (document, record, email)
- Every artifact lives in Google Workspace or Huly
- At each step, we show: "here's what the AI agent could do"
- The recruiter stays in control — agents assist, not replace
What the demo says to clients
- Your process must be digital before agents can help
- We help you design that digital workflow first
- Then we introduce agents at the right points
- The model and framework choices are yours — we advise
- Deployment is incremental, safe, and POPIA-compliant
2. The Recruitment Value Chain
1
JD & Brief
Human AI Assist
2
Source
AI Lead Human Review
4
Interview
Human Lead AI Prep
6
Offer
Human Lead AI Draft
3. Step-by-Step: Human Workflow + AI Opportunities
Step 1: JD & Client Brief
| Human Does | Where | AI Could | Agent |
|---|
| Client calls with a new role. Recruiter captures the brief. | Google Docs — Brief template in Drive/Clients/{client}/Vacancy_Briefs/ | Draft the JD from bullet points. Check for inclusivity flags (no "rockstar", age-coded language). | Ava (Acquire) |
| Create vacancy in ATS with description and salary range. | Huly ATS — create_vacancy via MCP or UI | Auto-create vacancy from the Doc brief. Link Drive URL as brief_drive_url. | Ava (Acquire) |
| Share brief with team. | Google Chat / Teams channel | Post summary to client channel with key requirements highlighted. | Elle (Engage) |
Step 2: Source Candidates
| Human Does | Where | AI Could | Agent |
|---|
| Search LinkedIn, job boards, internal database. | LinkedIn Recruiter, Indeed, existing talent pool | Scan talent pool against JD requirements. Rank candidates by semantic match. Flag passive candidates worth approaching. | Ava (Acquire) |
| Collect CVs — email, downloads, uploads. | Gmail → Google Drive /Candidates/Active/{name}/ | Parse incoming CVs automatically: extract name, skills, experience, salary. Create structured profile. | agilex-tools CV Parser |
| Create talent records in ATS. | Huly ATS — create_talent via MCP | Auto-create talent from parsed CV data. Attach Drive CV URL as Homepage channel. Deduplicate against existing talents. | Ava (Acquire) |
| Build longlist — link candidates to vacancy. | Huly ATS — create_application at Backlog stage | Auto-link high-match candidates to the vacancy. Set stage to Backlog. | Ava (Acquire) |
Step 3: Screen & Shortlist
| Human Does | Where | AI Could | Agent |
|---|
| Review CVs against requirements. Phone screen for fit. | Google Drive (CV) + phone call | Score candidates 0-100 against vacancy requirements. Classify into Tier A (interview), B (hold), C (reject). | Grant (Gauge) |
| Check salary expectations, notice period, availability. | Phone screen notes → Google Doc in candidate folder | Flag mismatches: salary above budget, notice period too long, location conflict. | Grant (Gauge) |
| Move shortlisted candidates to Screening stage. | Huly ATS — move to HR Interview | Auto-advance Tier A candidates. Hold Tier B for human review. | Grant (Gauge) |
| Send rejection emails to Tier C candidates. | Gmail | Draft constructive rejection emails with personalised feedback. Human reviews before sending. | Elle (Engage) |
Step 4: Interview
| Human Does | Where | AI Could | Agent |
|---|
| Schedule interviews with client panel. | Google Calendar + Gmail | Check FreeBusy for all attendees. Propose best slots. Create event with Google Meet link. Send prep pack. | Iris (Integrate) |
| Prepare candidate — send company info, panel profiles, tips. | Gmail with Drive links | Generate customised prep pack: company overview, interviewer LinkedIn summaries, role-specific talking points. | Iris (Integrate) |
| Conduct interview (client does this). | Google Meet / In-person | Generate structured interview questions based on CV gaps vs JD requirements. | Grant (Gauge) |
| Collect interview feedback from panel. | Google Sheets — scorecard template | Summarise feedback across interviewers. Flag consensus and disagreements. | Leo (Leverage) |
| Move to Technical Interview or next stage. | Huly ATS | Update pipeline stage based on feedback scores. | Grant (Gauge) |
Step 5: Assess & Reference Check
| Human Does | Where | AI Could | Agent |
|---|
| Request references from candidate. | Gmail | Send reference request emails from template. Track responses. | Elle (Engage) |
| Conduct reference calls. Record notes. | Phone + Google Doc in candidate folder | Generate reference check questionnaire tailored to the role. | Grant (Gauge) |
| Technical assessment (if applicable). | Client's assessment platform / take-home | Score technical submissions against rubric. Flag strengths and gaps. | Grant (Gauge) |
| Compile candidate summary for client. | Google Docs — summary template | Generate candidate brief: profile, scoring, interview feedback, references, recommendation. | Leo (Leverage) |
Step 6: Offer & Negotiation
| Human Does | Where | AI Could | Agent |
|---|
| Get package approval from client. | Gmail / phone | Generate salary benchmark report (market data for role + city + industry). | Leo (Leverage) |
| Draft offer letter. | Google Docs — offer template | Draft offer letter from template with package details, start date, conditions. Human reviews before sending. | Elle (Engage) |
| Present offer to candidate. Handle counter-offers. | Phone + Gmail | Draft negotiation talking points. Flag counter-offer risk indicators. | Elle (Engage) |
| Candidate accepts → move to Won. | Huly ATS — stage: Won | Auto-update pipeline. Trigger client notification. Generate placement report. | Elle + Leo |
Step 7: Placement & Onboarding
| Human Does | Where | AI Could | Agent |
|---|
| Confirm start date. Collect signed docs. | Gmail + Google Drive | Send onboarding checklist. Track document receipt. | Elle (Engage) |
| Invoice client. | Finance system / Google Sheets | Generate invoice from placement data (fee %, salary, dates). | Leo (Leverage) |
| Move candidate folder from Active → Placed. | Google Drive | Auto-archive candidate folder. Update pipeline tracker sheet. | Ava (Acquire) |
| Follow up at 30/60/90 days. | Google Calendar reminders | Schedule check-in reminders. Draft follow-up emails. | Elle (Engage) |
4. The Dev Environment
Working Stack
| Component | Role | Where |
|---|
| Google Workspace | Golden source of documents — CVs, briefs, contracts, scorecards, emails, calendar | agilex.co.za domain |
| Huly ATS | Pipeline tracking — talents, vacancies, applications, stages (metadata only) | Self-hosted (Docker) or huly.app |
| huly-recruit-mcp | MCP server — agents create/update ATS records programmatically | Node.js process (local or Cloud Run) |
| agilex-tools | CV parser API — PDF/DOCX → structured JSON via Workers AI | Cloudflare Pages |
| agilex-portal | Agent orchestrator — AI Gateway, model routing, D1 state | Cloudflare Worker |
| agilex-sim | Recruiter-first UI — Ava workflow demonstration | Cloudflare Pages |
| know.2nth.ai | Framework, model, and protocol reference for client conversations | 2nth.ai knowledge base |
Google Workspace Usage
| Service | Used For | Agent Integration |
|---|
| Google Drive | CVs, briefs, contracts, scorecards, offer letters | Ava uploads CVs, creates candidate folders. URLs stored in Huly as Homepage channels. |
| Google Docs | Vacancy briefs, candidate summaries, offer letters, reference notes | Agents draft from templates. Humans review and approve. |
| Google Sheets | Pipeline tracker, interview scorecards, placement reports | Leo reads/writes pipeline data. Scores aggregated automatically. |
| Google Calendar | Interview scheduling, follow-up reminders | Iris checks availability, creates events with Meet links. |
| Gmail | Candidate comms, client updates, reference requests | Elle drafts emails. Human reviews before sending. |
| Google Chat | Internal team communication, client channels | Agents post status updates to relevant channels. |
5. The Client Conversation
What we show
- The digital workflow — a real candidate moving through Google Workspace + Huly, step by step, with real documents and real emails.
- The AI opportunity map — at each step, we pause and show: "here's what the agent could do — parse this CV, score this candidate, draft this email, schedule this interview."
- The technology choices — using know.2nth.ai/explainers/agents/, we walk through: which frameworks (Google ADK, LangGraph, custom Workers), which models (Gemini, Claude, Llama), which protocols (A2A, MCP), which inference platforms (Workers AI, Ollama, Vertex AI).
- The deployment path — how to go from this demo to production: start with one agent (Ava), prove value, add more, scale safely.
What we say
"The A.G.I.L.E. agents don't replace your recruiters. They make your recruiters faster and more consistent. But for that to work, you need a digital process first — CVs in Drive, pipeline in an ATS, comms in email. Once that's in place, we can introduce agents at the right points. The model and framework choices from our knowledge base are part of the project with you — not a decision we've already made."
The know.2nth.ai reference
During the client conversation, the agents explainer provides:
- 7 frameworks evaluated — Google ADK, LangGraph, CrewAI, OpenAI, Anthropic, AutoGen, AgentForce — with trade-offs for each
- 3 protocols — A2A (agent discovery), MCP (tool access), Agent Skills (capability packaging)
- 6 model families — Gemini, Claude, GPT, Llama, Mistral, Gemma — with SA-specific notes (Ollama for POPIA, Vertex in africa-south1)
- 4 inference platforms — Workers AI (edge), Ollama (local), vLLM (production), OpenRouter (multi-model)
- 4 principles — agents as software (not prompt-tweaking), open standards, interop, data-driven decisions
This positions AgileX as an advisor, not a vendor. We don't push one stack — we help clients choose the right one for their context.
6. What to Build for June
| # | Deliverable | Status | Effort |
|---|
| 1 | Google Workspace setup — Drive folders, Docs templates, Sheets tracker, Calendar, Gmail for agenticX | To do | 2 hours |
| 2 | Huly workspace — running instance with MCP seed (1 vacancy, 5 talents across pipeline) | MCP ready | 30 min |
| 3 | Wire agilex-tools — deploy CV parser, connect to portal via CV_PARSER_URL | Ready to deploy | 30 min |
| 4 | Wire huly-recruit-mcp — HTTP wrapper so portal can call create_talent from Workers | Design done | 2 hours |
| 5 | Google OAuth — gate portal with @agilex.co.za login | Plan approved | 3 hours |
| 6 | One real candidate E2E — walk a real CV through all 7 steps in Workspace + Huly | After 1-5 | Half day |
| 7 | Demo recording / walkthrough — document the E2E with screenshots at each step | After 6 | Half day |
7. What Already Exists
| Asset | Repo | Status |
|---|
| CV parser (real PDF→JSON extraction) | agilex-tools | Working |
| Huly MCP server (create talent/vacancy/application) | huly-recruit-mcp | Working |
| MCP seed script (12-second workspace population) | huly-recruit-mcp | Working |
| AI Gateway (Gemini + Claude + Workers AI routing) | agilex-portal | Working |
| Recruiter-first simulator UI | agilex-sim | Working |
| A2A agent card | agilex-portal | Working |
| Google Workspace stubs (Drive, Gmail, Calendar, Sheets) | agilex-portal | Stubbed |
| Google OAuth plan | agilex-portal | Planned |
| Huly Integration Guide + MCP Spec | agilex-docs | Published |
| Agent Framework & Wiring Plan | agilex-docs | Published |
| Brand kit (light mode) | agilex-docs | Published |
agenticX June Demo Plan · agilex.co.za · 2nth.ai · know.2nth.ai/explainers/agents/