standalone
The Absorption Stack
How LLM capabilities reshaped knowledge work in five moves — and what’s left when the absorption finishes.
The absorption stack is the way to read what’s happened to knowledge work since late 2022 without getting lost in the parade of new tools. Five capability inflection points, each one absorbing a specific layer of human cognitive work — and each one freeing attention that immediately produced the training substrate for the next absorption.
You are not observing this from outside. You are inside the loop.
The Five Inflections
Nov 30, 2022 — Retrieval and Composition (ChatGPT)
What got absorbed: looking things up and producing first drafts.
What this created: the Prompt Crafter — practitioners who understood that the quality of the question shaped the quality of the answer.
What this dissolved: the entry-level researcher and drafter. Not the role entirely — the sole value proposition of being fast and fluent with information retrieval and text assembly.
Mar 14, 2023 — Multi-Step Reasoning (GPT-4)
What got absorbed: manual logical sequencing. The mental labor of connecting dot A to dot B to dot C, holding context across steps, producing a coherent argument or plan from scattered inputs.
What this created: the Framework Designer — practitioners who saw that the model could execute a logic chain if you handed it a good architecture to run inside.
What this dissolved: the analyst whose value was “connecting the dots.” Dot-connecting at human speed stopped being a premium service the day GPT-4 shipped.
May–Nov 2023 — Context Capacity and Packaging (Claude 100K + Custom GPTs)
What got absorbed: context curation. The skill of knowing what information to hold in mind and what to exclude — and the ability to deploy a curated context on demand.
What this created: the Platform Builder — practitioners who packaged their accumulated knowledge into custom GPTs, effectively turning their expertise into an always-on deployment.
What this dissolved: the “I have a methodology” moat. When the methodology lives in a system prompt and anyone with a ChatGPT Plus subscription can build the same deployment, the methodology itself stops being defensible.
Jun 2023–Nov 2024 — Tool Execution (Function Calling + MCP)
What got absorbed: manual interface work across SaaS tools. The labor of clicking, copy-pasting, navigating menus, translating outputs between systems.
What this created: the Workflow Architect — practitioners who understood how to wire tools together so the model could operate them without human intermediation.
What this dissolved: the operator whose value was “I know where to click.” Procedural familiarity with software interfaces stopped being a skill the moment models could read and operate interfaces themselves.
Jan 2025–Now — Task Orchestration (Operator, ChatGPT Agent, Claude Agents)
What is being absorbed: task orchestration itself. The cognitive work of breaking a goal into steps, deciding the sequence, monitoring progress, handling exceptions, and verifying completion.
What this is creating: the Agentic Manager — practitioners who set objectives, define quality gates, and review outputs, rather than executing the steps.
What this is dissolving: the executor whose value was “I can take this from start to finish.” End-to-end execution at human speed is not a premium service when agents run the same loop overnight.
The Loop You’re Inside
Each absorption freed attention. Freed attention immediately produced higher-quality training and workflow substrate — which accelerated the development and deployment of the next layer of absorption.
This is not a one-way compression from outside. Knowledge workers who adopted each tool actively contributed to the feedback loop: better prompts, clearer frameworks, sharper evaluations, richer workflow artifacts. The practitioners who engaged most deeply helped build the thing that absorbed the next layer of their work.
You are not a passive observer of this curve. If you are using these tools, you are helping to train what comes next.
Layer 6 — What’s Coming
The research signals converge on the same destination: continuous, self-verifying, memory-equipped, multi-agent operation. Systems that orchestrate themselves and improve through their own outputs. Not assistants. Not tools. Systems that run, catch their own errors, and update their own behavior.
Andrej Karpathy calls this the “loopy era” — when AI systems operate in continuous feedback loops rather than single-pass generations. Demis Hassabis points to 2026 as the year world models plus continual learning start producing genuinely novel scientific output. Dario Amodei describes it as a country of geniuses in a datacenter: not single superintelligence, but massive parallel intelligence operating at scale.
Layer 6 is not a tool you adopt. It’s an environment you operate inside.
Four Positioning Moves
1. Relocate before the absorption arrives, not after. The practitioner who positions at Layer 4 after Layer 5 has shipped is always behind. The pattern is visible enough now to read ahead. Watch where the next absorption is aimed, and start moving your value proposition before the landing.
2. Make your work legible, then immediately move on. Legibility — capturing your process, your frameworks, your decision logic — is not a moat. It’s a prerequisite for delegation. Make the current layer legible so you can hand it to a system, then move to the layer above.
3. Build the judgment muscle the next layer cannot encode. What does not get absorbed is the judgment formed by exposure to real cases, real consequences, and real feedback loops that no training set has seen. That accumulation is yours and it is structural — not because it’s secret, but because it’s embodied. It lives in how you read a situation, not in what you know about the domain.
4. Run the week test as ongoing practice. Each Friday: audit last week’s work by layer. Which blocks were Layer 2 work (drafting, researching)? Which were Layer 4 (operating tools, moving outputs between systems)? The absorption always arrives at the layer where most of your billable time still lives. The week test tells you where you’re exposed before the market does.
The Compression Is Not Purely Doom
Each absorption freed practitioners from work they often didn’t love. The analyst who spent 60% of their time building slide decks got that time back. The researcher who spent afternoons manually synthesizing sources got those afternoons back. The operator who lived in copy-paste hell between SaaS tools got out.
The question is not whether the compression will continue. It will. The question is whether you will be positioned at Layer 6 when it arrives — or still defending Layer 4 with a methodology document and a custom GPT.
The practitioners who are thriving inside the absorption are not the ones who moved fastest to adopt each new tool. They are the ones who used each tool to move themselves up the stack — spending less time at the layer the tool absorbed, more time at the layer above it.
That is still the move. It has been the move since November 30, 2022.