The anxiety around AI and jobs tends to produce two camps: people who think AI will replace everyone, and people who think it will change nothing important. Both are wrong. The real pattern is more specific and more actionable.
AI will not replace knowledge workers wholesale. It will first eliminate the least intellectually demanding parts of every knowledge worker's job — and that shift is already well underway.
The Augmentation Pattern
Think of every knowledge worker's job as a stack of tasks ordered by cognitive demand:
Bottom of the stack — Mechanical: data entry, formatting, copy-pasting between systems, generating boilerplate, scheduling, transcribing, basic summarisation.
Middle of the stack — Analytical: researching, synthesising information, drafting documents, answering standard questions, categorising, reviewing.
Top of the stack — Judgement: novel problem-solving, stakeholder negotiation, creative direction, ethical decisions, relationship management, context that requires deep domain expertise.
AI is eating from the bottom up. The bottom of the stack was already being nibbled at by automation tools. LLMs accelerated this massively and are now well into the middle. The top of the stack remains human — for now, and probably for a significant while.
Which Workflows Are Automating First
Legal and compliance
First-draft contract generation, clause extraction and comparison, compliance checklist verification, case law research, regulatory change monitoring. Large law firms are already running AI tools that handle first-pass due diligence at a fraction of the associate time. The analytical layer — junior associate work — is being compressed.
Finance and accounting
Variance analysis narration (explaining why the numbers changed), management reporting first drafts, invoice processing and reconciliation, audit sampling and anomaly detection. The quarterly close process that took a team two weeks is becoming a two-day AI-assisted workflow.
Software engineering
Code review, test generation, documentation, bug triage, PR description writing, refactoring suggestions, dependency audits. Senior engineers are already reporting they produce 40–60% more code with AI assistance. The junior-to-mid engineering tasks are being absorbed upward.
Customer operations
Ticket classification and routing, first-response drafting, FAQ handling, churn prediction, sentiment analysis at scale, escalation recommendations. The tier-1 and tier-2 support layers are compressing.
Marketing and content
SEO research, first-draft content, A/B test copy variants, social media scheduling, performance report narration, competitive analysis. Campaign execution — the implementation layer — is becoming almost entirely AI-assisted.
What This Means for Individuals
The knowledge workers who will thrive are those who move up their own stack — spending more time on the judgement layer and using AI to handle the mechanical and analytical layers efficiently.
The dangerous position: being valuable primarily for middle-of-the-stack work — analytical tasks that are reproducible and don't require unique context or relationships. This is where displacement risk is highest over the next 5–7 years.
The durable position: being the person who:
- Understands the domain deeply enough to evaluate AI output critically
- Holds the client or stakeholder relationships
- Makes the judgement calls that require context AI doesn't have
- Knows how to direct AI systems effectively to produce high-quality work
The second thing is a skill: AI direction. Knowing how to break a problem into subtasks, prompt effectively, evaluate outputs critically, and chain AI tools together is increasingly a core professional skill — not a niche technical one.
What This Means for Organisations
Companies that use AI to reduce headcount will get a short-term efficiency gain. Companies that use AI to give each person significantly more leverage will compound that advantage into capability.
The organisations building durable advantage are:
- Training existing staff to work with AI tools, not replacing them with AI
- Redesigning workflows around AI capabilities rather than bolting AI onto old processes
- Identifying which human judgement is genuinely irreplaceable vs which is habit
The implementation of this at scale requires someone who understands both the technology and the workflow design. If you're working through this for your organisation, let's talk.
The Timeline
The next 24 months will see significant change in the middle-of-the-stack layer. The analytical work that makes up the bulk of many junior and mid-level knowledge worker roles will require less human time per unit of output.
This is not a reason to panic. It is a reason to move now — to develop the AI fluency and judgement-layer skills that make you more valuable as the stack shifts, not less.