TL;DR (Working Ideas)
- Treat AI empathy as simulation until proven otherwise.
- Humans + Agents is the practical path for the next 5–7 years.
- Usage-based pricing and agent marketplaces will expand.
- Expect fastest gains in verifiable domains (code, maths, ops).
These are working ideas, not fixed beliefs. We’ll keep refining them as data, client results, and system telemetry evolve.
1️⃣ From Data to Wisdom — And Why Wisdom Stays Human
AI accelerates the path from data → information → knowledge. But wisdom — the mix of context, ethics, and cultural understanding — remains human-led. That’s the foundation of every NZGPTS workflow.
AI can simulate empathy. Understanding? That’s still human territory.
2️⃣ Where AI Already Excels
✅ Reasoning with clear feedback — code, maths, structured logic.
✅ Language ops — drafting, summarising, transforming.
✅ Operational speed — research, classification, triage, reporting.
Focus your automation efforts on tasks with tight feedback loops and clear success criteria.
3️⃣ The 5–7 Year Model: Humans + Agents
Humans: purpose, taste, empathy, relationships.
Agents: repetitive execution, multi-step workflows, API chores.
Governance: humans handle “why” and “what”; agents handle “how.”
Example Workflow Pairings
| Area | Human | Agent |
|---|---|---|
| Hiring | Defines values & selection criteria | Screens, enriches, books interviews |
| Customer Support | Defines tone & boundaries | Classifies, drafts, escalates |
| Reporting | Reviews trends, adds context | Compiles, formats, distributes |
4️⃣ Usage-Based Everything
- Shift from bundled plans → per-task usage.
- Agent marketplaces will emerge for on-demand specialists.
- Micro-payments (in cents or sats) will enable agent-to-agent economies.
Tip: Set daily budgets and approval thresholds early to avoid cost sprawl.
5️⃣ Vibe Coding vs AI Programming
- Vibe coding (0→1): natural language prototyping for front-end and ideation.
- AI programming (1→N): hardened code, security, tests, deployment.
Rule of thumb: prototype fast, ship safely.
6️⃣ NZGPTS Pilot Playbook for SMEs
- Inventory SOPs — find 3 repeatable processes.
- Agentise safely — automate the predictable steps; keep human approval.
- Right-size models — faster, smaller, cheaper when possible.
- Measure results — time saved, error rate, satisfaction.
- Iterate fast — kill or refactor in 30 days if no ROI.
7️⃣ Risks & Mitigations
| Risk | Mitigation |
|---|---|
| Hallucinations | Retrieval-Augmented Generation (RAG), confidence scores, review gates |
| Security | Least privilege, handcrafted auth, peer review |
| Cost Drift | Usage telemetry, daily caps, auto-downgrades |
| Cultural Missteps | Human sign-off for brand & Te Ao Māori tone |
8️⃣ How We’ll Keep Updating These Ideas
Our stance: “ideas are tools, not trophies.”
As telemetry, case studies, and client data evolve, so do our frameworks.
We’ll keep this roadmap adaptive — just like the systems we build.
🚀 Ready to Pilot Humans + Agents?
Book a 30-minute fit call — we’ll map 3 SOPs → 3 agents, confirm ROI metrics, and deploy your first pilot safely within 14 days.
Amy Ferguson
Founder, NZGPTS
Making AI genuinely useful for NZ businesses — with the mana left in human hands.