01

Global Minds 2026

AI isn't
coming.
It's here.

A practical briefing on what's real, what's changing,
and what to do next. Teddy Pejoski

02

Compute Conference, SF · March 9 2026

The toy became
the tool.

50% of Mintlify's docs traffic is now
AI agents — up from 15% one year ago
Dec 2024

342K AI views out of 2.3M total — 15% of traffic

Dec 2025

~10M AI views out of 20.8M total — 50% of traffic

End of 2026 (projected)

90% agents · 10% humans — Han Wang, Mintlify CEO

03

Mental model

Most people are
missing the stack.

Models

OpenAI, Anthropic, Google, Meta — building the intelligence

Products

ChatGPT, Claude, Gemini — the interfaces you actually use

Agents & Harnesses

Systems that plan, act, use tools, and complete work

"The harness is the new product — not the model, not the agent." — Harrison Chase, LangChain

04

Definitions

Chatbot. Agent.
Workflow.
Not the same.

Chatbot

Answers your prompt

Great for thinking, drafting, and exploring ideas in conversation.

Agent

Takes a goal, uses tools

Completes multi-step work autonomously. Needs guardrails and review.

Workflow

Repeatable system

Durable value lives in the loop — not just the prompt alone.

05

Judgment

Useful now.
Immature
elsewhere.

Already real
  • Drafting, summarising, translation, research support
  • Coding help, debugging, technical exploration
  • Bounded workflows where a human reviews the output
Still maturing
  • Unattended automation in high-stakes settings
  • Polished demos presented as stable products
  • Agents paying $0.10 for API access — exciting, but early
    — Aaron Levie, Box · Compute Conference
06

How work changes

AI hits tasks.
Tasks stack
into jobs.

Lower exposure

Physical / situational

Trades, on-site service, unpredictable real-world work.

Medium exposure

Relational / judgment

Healthcare, management, client trust, complex negotiations.

Higher exposure

Digital / repeatable

Research, writing, legal analysis, recruiting, finance, software.

07

The new knowledge work

Work becomes
workflow
design.

Tool Footprint — TF(time, tokens)

Agents pick the fastest, cheapest tools. If your tool is slow or noisy, they switch. The lowest footprint wins.

Judgment still matters

The higher the stakes, the more human review matters. You're the editor, not just the prompter.

You don't need to code

But fluency in how digital work gets structured is increasingly non-optional.

08

Personal adaptation

Four levels.
Most people are
stuck at one.

↑ Leverage
L4

AI-native

Builds with AI, multiplies output, designs the workflows others follow.

L3

Regular operator

AI in daily work — defined prompts, real output, consistent routine.

L2

Casual user

ChatGPT occasionally. No workflow integration. Random prompts.

L1

Non-user

Hasn't tried AI seriously. Skeptical, overwhelmed, or just unaware.

Gary Tan AI Autonomy Ladder (developer reference)

via Gary Tan · L1–L8 for devs

09

This month

Pick one.
Do it
this month.

Writing

Drafts & outreach

Emails, proposals, summaries — let AI draft, you edit.

Research

Briefings & synthesis

Turn 20 tabs into one sharp brief in 10 minutes.

Code

Automate the boring

Spreadsheet logic, data cleanup, small scripts — no CS degree needed.

Ops

Meeting notes & plans

Record → transcript → summary → action items. One loop.

10

Global Macedonians

AI can strengthen
communities.
If you design
it that way.

Memory

Stop losing institutional knowledge when people move on.

Coordination

Cut repetitive admin. Let leaders focus on what matters.

Shared intelligence

Help members get answers fast — without replacing human leadership.

Global Macedonians community
11

By 2027

AI becomes part
of the environment.

B2A replaces B2B/B2C

Business to Agent

Agents are already paying for access, making decisions, switching tools — without a human in the loop.

New metrics

DAA / MAA

Daily Active Agents. Monthly Active Agents. Alongside DAU / MAU — the playbook is being rewritten.

IAP replaces ICP

Ideal Agent Profile

Who are your target agents? What tools can they use? What context do they retain? What model do they run on?

12

Closing actions

Start here.

01 — Use it regularly

Build the reflex

Occasional experiments don't build fluency. Daily use does.

02 — Improve one workflow

Find a recurring task

Pick one thing you do every week. Make it faster or sharper this month.

03 — Review the output

You're the editor

The skill isn't just generation — it's evaluation. Stay in the loop.

13

Q&A

"The future belongs to people who learn to work with these systems."
Clarity

Understand the stack.

Judgment

Separate signal from noise.

Action

Improve one real workflow.

Teddy Pejoski · Global Minds 2026