Trust and quality notes
- Last updated
- October 3, 2026
Adding AI to an old process can improve the process without changing what it is for. That may be useful, but Sumeet Singh argues that it will not produce the most important new businesses. His thesis is that founders should stop treating AI as a layer on existing workflows and ask what kinds of work were technically impossible before.[1]
I read this as a view of humanity after automation centered on invention rather than replacement. The opportunity is not just to make familiar tasks faster. It is to redesign how people act around capabilities that earlier software did not provide.
The source's central argument
Singh says the era of building specialist software through the framework of the previous decade has ended. He expects businesses that begin with an existing workflow and simply “AI-ify” it to fail, while those that use the distinctive properties of models to create new workflows will survive.[1]
He calls the latter post-skeuomorphic apps. Skeuomorphism, in his explanation, is the assumption that a new technology should resemble what came before. Early mobile apps often copied physical objects or existing processes rather than exploring the phone's unique abilities.
The breakthrough mobile companies did something different. Singh uses Uber as his central example. It did not merely digitize a taxi dispatcher's desk. It built around the fact that many people carried connected devices that knew their location. DoorDash and Instacart followed the broader pattern of turning the phone into what investor Matt Cohler described as a remote control for life.[1]
Singh believes AI has reached a comparable inflection point. The important question is not how to insert a model into today's steps, but what work can now exist because of the model. He is candid that we do not yet know what the winning workflows will look like.
Why familiar interfaces can hide a narrow imagination
New technology is often introduced through familiar forms because familiarity reduces confusion. That can be a sensible bridge. Singh's warning is about mistaking the bridge for the destination.
If a team begins with a fixed sequence of steps, it may use AI only to accelerate one step at a time. The result could be faster drafting, classification, or search while the underlying assumptions remain untouched. Who initiates the work? What information must be collected first? Why are handoffs arranged in this order? Which parts exist only because earlier software could not understand unstructured input? Those questions stay unanswered.
A post-skeuomorphic approach starts with the technology's new properties. Singh does not enumerate those properties or provide a recipe. He asks founders to examine what models uniquely make possible and imagine a workflow native to those capabilities.
The human role is to ask a better opening question
Singh's preferred question is simple: “What becomes possible now?”[1] That moves human contribution toward problem framing and institutional imagination. A model may perform parts of a new workflow, but people still decide which possibilities are worth pursuing, whose needs matter, and what a better arrangement should accomplish.
This matters for readers new to AI because the easiest starting point is often a visible annoyance in an existing process. There is nothing wrong with removing that annoyance. But Singh's argument suggests a second pass. After improving the current step, ask whether the step should exist at all. A process built around forms, queues, and specialist interfaces may reflect technical constraints that no longer apply.
Humanity after automation, in this account, is not passive. People are not simply waiting to learn which tasks remain. They are designing new forms of work around a changed medium.
What is argument and what is interpretation
Singh explicitly predicts that companies which merely add AI to existing workflows will lose to companies that invent workflows made possible by AI. He grounds this claim in an analogy to mobile computing and names post-skeuomorphic applications as the desired category.[1]
My interpretation is that teams can explore this thesis with two maps. The first map shows today's workflow exactly as it operates. The second begins with the desired outcome and ignores the current sequence. For each requirement, ask whether it exists because the user needs it, because an institution requires it, or because older software imposed it. The third category is the richest candidate for redesign.
This method is not in Singh's short article. It is a practical extension of his argument. It also needs a caution: not every established step is obsolete. Some steps encode accountability, safety, consent, or a valuable human conversation. Removing friction without understanding its purpose can damage the outcome.
A second interpretation is that novelty alone is not enough. Singh argues for new workflows, not decorative new interfaces. A genuinely new workflow should change what a person can accomplish, not merely make the experience feel futuristic. The source does not claim that any use of an old pattern is doomed; its target is a strategy that never moves beyond old assumptions.
How to search for a native AI workflow
Begin with a result people care about, not with a model feature. Describe the result without naming the current software. Then identify the information, decisions, and permissions required to reach it.
Next, mark the constraints created by older tools. Does a person translate natural language into rigid fields because the system cannot interpret a request? Are several specialists handing off work because no shared tool can maintain context? Does the workflow require a separate interface for each narrow action? These questions can reveal where the old form is driving the process.
Now consider what a model uniquely changes in this particular workflow. Describe the new interaction or result as a hypothesis, then test it rather than treating novelty as a guaranteed improvement. Compare the whole experience, including accuracy, clarity, control, and recovery when the system is wrong.
Finally, preserve the human purposes hidden inside the old process. A meeting may carry trust as well as information. A review may create responsibility as well as approval. Singh asks us to leave inherited forms behind, but a thoughtful redesign carries their legitimate functions forward.
Invention is a responsibility, not a slogan
Singh's thesis is energizing because it refuses to define AI by the tasks people already perform. It is also demanding because nobody can copy the answer from the previous software cycle. The future workflow has to be discovered through observation, design, and real use.
Do not confuse automation with imagination. Making an inherited process faster can create immediate value. Creating a process that could not exist before may change what people believe software is for. The strongest work will know which goal it is pursuing and judge itself honestly.
If you want to explore a workflow beyond simply adding AI to old steps, Agentic Workers can help you turn a real outcome into a testable new way of working.