The first great interface promise was WYSIWYG: What You See Is What You Get. The screen showed you a document or page that closely resembled the eventual output. This made direct manipulation possible for people who did not want to learn formatting commands or typesetting systems. But the screen did not research a market, decide whom the message was for, commission a photograph, design the customer journey, write the code, test the result and publish it. Someone — usually several people — still had to carry out that production chain.
The newer promise we call WYTiWYG: What You Think Is What You Get. Its distinctive word is THINK. A user describes the desired state, not every intermediate action. An AI system can then help clarify the question, assemble context, plan the work, use tools and produce a reviewable result. WYSIWYG shortened the distance between execution and appearance. WYTiWYG shortens the distance between intention and execution. The change is not merely a better text box; it shifts the level at which people direct computers.
SEE made the output tangible, not effortless
Early visual editors such as Bravo on the Xerox Alto gave users direct control over formatted text; Computer History Museum’s profile of Charles Simonyi places Bravo among the pioneering WYSIWYG editors. A user could change a font or layout and immediately inspect the effect. This removed an important translation step between commands and output.
What it did not remove was the work before the screen. A campaign still needed a brief, an audience, evidence, copy, visual assets, approval, implementation and quality assurance. The editor let a designer see the artefact while working on it. It did not infer whether the artefact addressed the right problem. That is why SEE is the decisive word in the old formula.
THINK begins with a result, and sometimes with a question
Imagine telling a system: “Build a product page for independent pharmacy owners. Explain the programme, show three substantiated outcomes, collect enquiries and meet technical SEO requirements.” That is not a sequence of clicks. It is an intention with an audience, constraints and acceptance criteria. A capable agent can propose the structure, identify missing evidence, prepare copy, modify code, run checks and present a draft for approval.
The deeper shift appears when the user has no complete brief. A conversational system can ask: Who will make a decision after reading this page? Which figures are verified? What claim would be too risky to publish? In that sense, AI participates not only in the GET but in the THINK. Google’s Deep Research illustrates this with a reviewable research plan that can evolve as sources are explored. Anthropic’s account of agent design distinguishes a predefined workflow from an agent that chooses steps and tools while pursuing a goal.
Claude, Gemini, Copilot and Perplexity occupy different parts of the chain
Claude Code and other coding agents can inspect a repository, modify multiple files, run tests and iterate. An Anthropic–Rakuten case study describes a seven-hour Claude Code run on a task in the vLLM project and reports a reduction in average feature delivery time from 24 to five days in the wider deployment. These are company-reported results, not an independent estimate of what every engineering team can achieve. Their significance lies in the continuity of the work: a single task can move across analysis, implementation and checks without repeated hand-offs.
Gemini Deep Research starts with a research question and turns it into a plan and sourced synthesis. Perplexity Labs expands a single request toward reports, spreadsheets, dashboards and simple applications. GitHub Copilot and Cursor can assist in coding and, in agentic modes, carry out parts of a development task. In visual production, Firefly, Midjourney, Canva and Runway move from the brief or reference toward image and video variants. ChatGPT is part of this landscape, but no single product owns the transition. What matters is how much of the path between an intention and an artefact a system can traverse reliably.
Faster generation does not always mean faster work
An experiment on GitHub Copilot found that participants completed a bounded JavaScript task 55.8% faster with assistance. In a very different setting, a METR study of 16 experienced open-source developers across 246 real tasks found that using early-2025 AI tools made completion 19% slower, despite developers’ perception that they were faster. The studies need not cancel each other out. They describe different tasks, constraints and verification costs. Generation can be cheap while reading, checking and repairing a plausible but imperfect result is expensive.
There is evidence of a broader shift in team work too. In a field experiment involving 776 Procter & Gamble professionals, an individual working with AI produced solutions comparable in quality to those of two-person teams without it. AI also helped bridge technical and commercial perspectives. This is not proof that organisations no longer need teams; it is evidence that some of the cognitive benefits of a team can reach an individual earlier in the process.
The bottleneck moves from production to judgement
When AI can help formulate the question, conduct research, generate materials, build a page and test code, the scarce resource is less often the first draft. It becomes the ability to choose the right problem, provide credible context and reject a polished answer that is wrong. In a Microsoft Research and Carnegie Mellon study covering 319 workers and 936 use cases, critical thinking shifted toward verification, integration and oversight; higher confidence in AI correlated with less critical scrutiny.
So “just think and it happens” is a useful description of a bounded digital workflow, not a licence to remove accountability. The practical sequence is: name the desired change, let AI challenge unclear assumptions, approve the plan, delegate work within defined permissions, and inspect the evidence and result. SEE returns at the end as quality control. WYTiWYG is not a promise that every thought becomes a flawless product. It is a promise that a well-expressed, well-tested intention can now mobilise capabilities that once required a much longer production line.