Single-threaded by design
You can only truly "hold" one thing in attention at a time. Everything else is a context swap. AI is valuable when it lets you stay in one lane longer.
The Paper
Constraint Migration in AI-Native Work
A framework for understanding how AI changes work. It doesn't remove constraints, it relocates them. Built on operations management, lean thinking, and Goldratt's Theory of Constraints.
The Argument
Your brain is single-threaded. Most "productivity" is just you reloading context all day: who is this lead, what did we agree, what's next, what's the tone, where's the doc, what did I promise. AI doesn't replace judgment. It replaces reloading.
$30 work vs $300 work
Locating files, formatting, checking updates, scheduling: that's $30/hour execution. Direction, decision, judgment: that's $300/hour work. Stop running execution on your brain.
Context switching is a tax
Each task switch costs ~23 minutes of refocus (Gloria Mark research). AI handles the context-reloading so you can stay in flow on what matters.
Orchestrator, not executor
The shift: you direct systems instead of doing tasks. AI agents do the work. You decide what matters.
The 5 Constraints of AI-Native Work
AI doesn't remove constraints, it relocates them. If you don't know where the new bottlenecks are, you've just bought tools. You haven't changed how you work.
1
Context
AI lacks the information it needs. Build context libraries and structured knowledge. Don't expect AI to guess.
2
Control
Unclear permissions prevent AI autonomy. Define approval thresholds, policy boundaries, and escalation points.
3
Confidence
Miscalibrated trust: checking everything or checking nothing. Spot-check routine work, full audit for anything risky.
4
Coordination
Poor handoffs between AI, tools, and humans kill flow. Clean workflow architecture and standardised output formats solve this.
5
Capacity
Human review bandwidth becomes the new bottleneck. Prioritise reviews, increase confidence thresholds where safe, and hire reviewers if needed.
Most organisations are stuck on one or two of these. The question is which ones.
Decision Framework: Use, Customise, or Build?
Most people jump to building when they should be using what already exists. This decision tree saves you weeks.
Use existing tools
1–2 hours
If an existing tool solves 80%+ of the problem, use it. ChatGPT Apps, Claude Desktop, Zapier, AskYourPDF, Canva. This covers most people.
Customise
2–5 hours
If you can configure it, do that. Custom GPTs, Agent Skills (SKILL.md), Claude Projects. No code required. Handles 80% of remaining cases.
Build custom
Calculate ROI first
Only build if nothing else works AND the ROI justifies it. MCP servers, custom agents, API integrations. This is where engineering meets domain expertise.
Theoretical Lineage
This framework applies Goldratt's Theory of Constraints to AI-enabled work.
Goldratt (1984)
Theory of Constraints: every system has one binding constraint. Optimising anything else is waste. Identify it, exploit it, subordinate everything else to it.
→
Lean & Operations
When you relieve one constraint, it doesn't disappear. It migrates. The bottleneck in a manufacturing line moves from machine to machine as you improve each one.
→
AI-Native Work
Same thing is happening with AI. Execution was the constraint. AI relieved it. Now Context, Control, Confidence, Coordination, and Capacity are the bottlenecks, and most people haven't noticed the shift.
Operations management studied at NTU under Prof. Roy Stratton and Prof. John Disney (HS2), who worked directly in constraint theory. This paper applies that discipline to AI workflow design.