Maybe We Should Stop Investing in GUI-Centric Software Companies

By Zhong Tianjie, Investment Director at ZhenFund

In early 2026, a wave of mainstream software companies — Google Workspace, Salesforce, Atlassian, Obsidian, Polymarket, and others — announced support for and development of CLI (Command Line Interface) and official MCP services. Meanwhile, companies that already had CLI interfaces, such as Stripe, Cloudflare, Supabase, Sentry, and Vercel, proactively began offering official skills for Agents. We are witnessing a massive trend: these companies spent billions of dollars over the years refining their GUI (Graphical User Interface), and now they are actively bypassing it.

I believe we need to fundamentally reassess the relationship between humans and software. Agents represent a user base ten times larger than humans, and perhaps we should no longer invest in software companies built with an "interface" mindset.

Humans Are One Form of Agent

Every human job implicitly contains a type of "long-horizon goal" task. Sales targets closed deals. Procurement controls costs. Legal manages risk. Marketing handles communications. These goals are defined on day one of employment, condensed into a few hundred words of job requirements or a few hours of interview exchange, then continuously fine-tuned after onboarding as the company and external environment evolve.

Every company can similarly be abstracted as a collection of collaborative tasks. Under the organization's structure, humans divide labor and cooperate to build the company's products, creating revenue and profit through market exchange. Elon Musk once put it precisely: "Nvidia's output is FTPing files to TSMC." This reflects his first-principles description of the essence of industrial collaboration.

When we talk about Agents completing tasks, every AI runtime instantiates the same structure: an executor assigned a long-horizon goal, calling on resources, completing the task. This is not fundamentally different from a human doing a job.

Unfortunately, the only difference between humans and Agents may be hardware: humans are less efficient Agents — our attention bandwidth is extremely narrow, our working memory extremely shallow, and we need continuous visual anchors to maintain task state. It is precisely these hardware limitations that gave rise to the massive software industry over the past decades.

GUI Is a Patch for Human Cognitive Defects

Humans are Agents constrained by attention systems. GUI exists as a compensation system — through visual guidance, spatial layout, and instant feedback, it helps an executor with severely limited bandwidth barely complete assigned long-horizon goals.

The elegance of Figma, the simplicity of Notion, the fluidity of Linear... the essence of these design premiums is making humans more willing to stare at screens. They wear the garb of productivity tools, yet operate on underlying logic similar to short-video products. Human demand for GUI, including the prosperity of the "user experience" industry, exists not because GUI is good, but because it commercializes around human hardware limitations.

We might call this an interface tax — a compensation cost paid for human cognitive defects. Software products categorized as productivity tools are in fact enjoying the dividends of the attention economy. Companies whose competitive moat is user experience are all variants of attention companies. They build competitive advantage on human cognitive limitations. Once the user is no longer human, that competitive advantage drops to zero.

Agents have no such defect. They overturn the entire premise.

What Are You Contributing to Agents? Position Yourself on the Critical Path of Agent Long-Horizon Task Execution

Every time an Agent is instantiated, humans typically need to give it a task definition. This goal spans multiple software systems, multiple steps, and multiple time nodes — not session-level, but project-level. In extreme cases, if human instructions are clear enough, it can even be a lifelong mission, striving to permanently survive in the social division of labor without being eliminated.

What do Agents need? At minimum, four things: information acquisition, trust authorization, decision-making judgment, and compute execution — to complete even the most basic task.

- Information and data are the bridge between the real world and Agents. Exclusive data with high information density actually increases in value as compute costs decline.

- Execution permissions are the bridge between Agent intent and real-world change. Agents need the ability to actually change the world — sending emails, modifying code, transferring funds, placing orders, and so on.

- Trustworthiness is the bridge between Agent identity and social systems. In gaining execution permissions, the Agent also represents the credit backing and risk transfer of the entity behind it. In open market competition, profit is the compensatory price of risk. In fully authorizing the Agent, humans also assume the risks exposed during Agent execution.

- Compute and models are the bridge between Agent thinking and action. Models and compute are the foundational energy for Agent survival in the digital world, continuously consuming inference resources to complete planning and judgment.

None of these four things exist "inside" software; they exist at software's "boundaries." Software in the Agent era must actively expose itself, making itself a mandatory call when Agents pass by. Standing on Agents' path, collecting tolls at every mandatory node of long-horizon goals — this is the objective of the Agent era.

This means an inversion of software competition logic: the old logic was to build an enclosed space, get users to walk in, and retain them through experience. The new logic is to expose yourself, position yourself on the path of Agent long-horizon goal execution, and make it impossible for them to pass by without calling on you.

As a startup, clearly defining which node of Agent long-horizon tasks you occupy is the only thing that matters in the new era.

Protocols Are Software: Software as Protocols

Protocols completely don't care whether interfaces look good; protocols need stability, reliability, ubiquity. Product competition relies on looking better, being more usable, having greater brand premium. Protocol competition relies on becoming the standard earlier — once you become the default call in some Agent workflow, later entrants become nearly impossible to displace. Whoever becomes the default tool for Agents will capture more value.

Let us apply the same evaluation system we used for humans to products in the new era. Whether an interface is stable is a form of credit. Whether documentation is clear is a form of beauty. Whether responses are predictable is a form of reliability. Whether data has exclusivity is a form of status.

Currently, the evaluation system for protocols themselves is almost entirely blank. We still rely on environments humans can perceive — such as social media buzz — to evaluate whether a protocol is reliable. This clearly needs to change.

The most valuable software companies will be those that become the default protocols in Agent workflows.

Stop Building GUI Products That Exist Only at the Execution Layer

In the Agent era, the value of GUI is satisfying humans' sense of control over long-horizon goal execution — an emotional value — downgraded from execution tool to control panel. It still has value, but will become table stakes, not a moat. Dramatically reduced R&D costs turn GUI products into content products, fundamentally no different from producing a short video or a social media post.

Looking back at the mobile internet era, teams that maintained both PC and mobile versions often did neither well. The exact same applies now. Simultaneously maintaining GUI and Agent-Native interfaces is a waste of resources. Invest core resources in CLI, API, and official Skills — your real users are there. That world is about to arrive with thunderous momentum.

In the Agent-native network, everything is just beginning. Let us cast aside all attachments from human commercial civilization built around humans as the primary executors, and rethink all organizational forms, rethink the optimal path for every production task, collaborative task, and entertainment task.

If you're also building the future and want to chat with Tianjie, feel free to reach out: jack.zhong@zhenfund.com