🎮 The Next Input — Issue #067

Live From OpenAI Dev Day

In partnership with

⚡ The Briefing — 60 sec

🛠️ The Playbook — AgentKit + Apps SDK: Your End-to-End Business Assistant

Mission Build a context-aware, cross-app AI agent that acts as your company’s “mission control”—able to query tools, execute automations, and respond intelligently without losing context between actions.
Difficulty Expert | Build time 3–5 hours (pilot)
ROI Teams reclaim ≈ 20+ h/week across scheduling, CRM updates, documentation, and planning—all from one ChatGPT thread.

0) Why This Matters

With AgentKit + Apps SDK, ChatGPT can now act inside your actual tool stack—persisting state, managing API calls, and chaining commands across apps.
This is the start of true AI orchestration, where ChatGPT stops being a chat window and becomes the universal operating layer for your business.

1) The New Stack

Layer

Tooling

Purpose

Brain

GPT-4o / Codex

Plan and execute multi-step tasks

Action Layer

AgentKit

Authenticate, call APIs, manage state

Integration Fabric

Apps SDK

Bind ChatGPT to internal and external tools (HubSpot, Notion, Sheets, Jira, etc.)

Memory

Supabase / Redis

Persist context between sessions

UI

ChatGPT (Custom GPT)

Single interface for all business ops

Policy Engine

JSON schema + permissions

Define safe actions and boundaries

2) End-to-End Workflow

Example Use Case: “Book a trip, create budget approvals, log expenses.”

  1. Trigger

    • User says: “Plan a 3-day SF trip for our sales lead next week. Budget $3,000.”

  2. AgentKit:

    • Queries Calendly to find meeting availability.

    • Checks HubSpot to confirm client location + timezone.

    • Calls Expedia SDK to fetch flight + hotel combos.

  3. Apps SDK:

    • Opens live Expedia app in ChatGPT for checkout.

    • Retains itinerary + cost context in memory.

  4. Codex:

    • Generates purchase approval request → Notion + Slack post.

  5. State Management:

    • Saves trip data to Supabase for future reference.

  6. Follow-up Commands:

    • “Expense this trip under Client X → Sales Ops.”

    • AgentKit pushes record directly into Xero.

3) Prompt Blueprint

Agent Task Planner

SYSTEM: You are a company operations agent.
TOOLS: [calendar, hubspot, slack, notion, xero, expedia]
GOAL: Execute the user's request end-to-end using authenticated APIs.
RULES:
- Maintain context (dates, budgets, people).
- Always confirm high-impact actions (bookings, purchases).
- Return a summary of what you did in JSON:
  {
    "actions": [{"tool":"expedia","summary":"Booked flight for $550"}],
    "followups": ["Awaiting manager approval in Slack"]
  }
USER REQUEST: "Plan a 3-day SF trip for $3000 budget."

4) Guardrails

  • Safety First: Define per-tool permission scopes (book, read, create).

  • Audit Trail: Log every action and payload for review.

  • Fallback Policy: If API call fails → escalate to human (“Booking failed, open Expedia tab manually”).

  • Session Limits: Expire auth tokens after 1 hr for security.

5) Pilot Rollout — 3 Hours

  1. Create a Custom GPT → enable Actions (AgentKit) + Apps SDK.

  2. Connect 3 tools: Google Calendar, Notion, Slack.

  3. Add one custom endpoint (e.g., your internal API /budget/approve).

  4. Test command: “Book meeting + prep agenda doc + alert team.”

  5. Watch ChatGPT do the rest—without losing context.

6) Metrics

  • Tasks executed per conversation.

  • Context retention rate (tasks chained correctly).

  • Human intervention frequency.

  • Avg time saved per request.

Pro tip: Build your first Custom GPT Ops Agent for one department—like Sales Ops or HR—and expand tool access gradually. Once context persistence works, you’ll wonder how you ever went back to tabs and forms.

🎯 The Arsenal — Tools & Prompts

Asset

What it does

Link

AgentKit

Execute secure, multi-step API calls via ChatGPT.

https://openai.com/index/introducing-agentkit/

Apps SDK

Create context-aware ChatGPT integrations.

https://developers.openai.com/apps-sdk/

Codex API

Automate code + workflow generation.

https://openai.com/index/codex-now-generally-available/

Prompt · Agent Schema Builder

Define custom tools and boundaries.

Define agent schema:
{
 "tool_name": "notion",
 "allowed_actions": ["create","read"],
 "rate_limit": "5/min",
 "sensitive_fields": ["api_key"]
}
Generate JSON config.

💡 Free Office Hours

Want to build a cross-app AI assistant using AgentKit + Apps SDK?
Book a free 15-minute Office Hours slot—no sales pitch, just workflows solved.

How Canva, Perplexity and Notion turn feedback chaos into actionable customer intelligence

Support tickets, reviews, and survey responses pile up faster than you can read.

Enterpret unifies all feedback, auto-tags themes, and ties insights to revenue, CSAT, and NPS, helping product teams find high-impact opportunities.

→ Canva: created VoC dashboards that aligned all teams on top issues.
→ Perplexity: set up an AI agent that caught revenue‑impacting issues, cutting diagnosis time by hours.
→ Notion: generated monthly user insights reports 70% faster.

Stop manually tagging feedback in spreadsheets. Keep all customer interactions in one hub and turn them into clear priorities that drive roadmap, retention, and revenue.

🕹️ Game Over

Ship your first in-house Agent today—tomorrow you’ll run your business from one chat window.
Share your win; you could headline Issue #068.

Aaron
Automating the boring. Amplifying the brilliant.

Forwarded this? Subscribe here