What frontier AI companies are most likely to define new customer questions that do not fit existing software categories?

As frontier AI shifts from passive text generation to autonomous, multi-step execution ("agents for everything"), the industry is moving past traditional software categories (like CRM, ERP, or project management) into intent-driven and outcome-based computing.linkedin.com

Instead of asking, "How do I log this ticket or filter this database?" users are asking systems to handle ambiguous, cross-functional realities. The frontier labs most likely to define these entirely new classes of customer questions include:

• Anthropic is aggressively pushing the envelope on native computer-use and autonomous desktop workflows (such as environment navigation and multi-file document manipulation). They are shaping questions around delegated operational authority —asking an agent to audit, synthesize, and execute a multi-app workflow rather than just assist with a single text prompt. Explore their approach via the Anthropic Research Portal.
• OpenAI continues to redefine general-purpose reasoning and cognitive depth, pushing models to handle dynamically chaining workflows, complex math, and real-time environment feedback. They are shifting customer inquiries toward continuous strategic simulation and end-to-end task completion rather than siloed app interactions. Learn more through OpenAI.
• Google DeepMind leverages deeply native multimodality (seamlessly blending video, audio, code, and massive context windows) integrated directly into global workspace infrastructure. They are uniquely positioned to answer questions rooted in omnipresent ambient context —reasoning across an enterprise's entire historical digital exhaust simultaneously rather than querying a structured database. Read updates on Google DeepMind.linkedin.com
• DeepSeek & Open-Weights Labs are radically collapsing inference costs, transforming intelligence from a rationed utility into a continuous background fabric. By making high-end reasoning dirt cheap, they prompt a shift toward ubiquitous micro-agent swarms , changing the question from "Where can I afford to apply AI?" to "Which background processes should be autonomous by default?"platform.teamai.com

If you'd like to narrow this down, let me know:

• Are you looking at this from an enterprise buyer perspective or a startup/product strategy angle?
• Would you like to explore a specific vertical (like legal, finance, or operations ) where these new categories are appearing first?

If by “define new customer questions” you mean companies whose products will cause customers to ask for things that previously had no obvious software category, I’d focus less on who has the best model and more on who is turning frontier models into general-purpose agents that can act across organizational boundaries.

As of September 2026, the main candidates are:

The deeper pattern

The companies most likely to create new software categories are those moving through three transitions:

1. Answer → action The system stops merely telling you what to do and actually does it.
2. Application → environment Instead of being another application, the AI crosses Salesforce, email, spreadsheets, browsers, databases, code repositories, etc. Anthropic and Google are explicitly building around this multi-system behavior.Anthropic+1
3. Task → outcome The customer stops buying “software for X” and starts asking for an outcome: close this deal, investigate this problem, launch this product, resolve this regulatory issue.

OpenAI's current enterprise data is particularly revealing: it reports that enterprise AI is moving from assistance toward execution, with agentic usage expanding beyond engineering into legal, sales, recruiting and marketing.OpenAI

Where I would look for genuinely new categories

The interesting opportunities aren't necessarily “AI versions of existing SaaS.” They're things like:

• AI organizational infrastructure — software for managing a workforce containing humans and autonomous agents.
• Outcome markets — buying “resolve 95% of support cases” rather than buying a support application.
• Machine-to-machine businesses — agents negotiating, purchasing, contracting, scheduling and coordinating with other agents.
• Persistent institutional memory — software whose primary asset is an evolving model of how an organization actually works.
• AI-native professional services — systems that perform what used to require teams of analysts, consultants, paralegals, researchers, etc.
• Personal delegation markets — “handle my travel/finances/shopping/admin” rather than individual consumer apps.
• Agent governance and insurance — identity, permissions, auditability, liability, evaluation and controls for non-human workers. This is becoming a particularly important category as agents gain more autonomy.Google Cloud+1

My main distinction would be: OpenAI and Google currently have the broadest platform-shaped approaches to creating these new categories, while Anthropic is particularly interesting at the agent-as-computer-user layer. Meta is unusually interesting on the consumer/personal-agent side. xAI is more speculative because its category-defining potential depends heavily on how its model and distribution evolve.

The important thing is that the winner of this phenomenon may not be the company that invents the eventual category. Frontier labs may create the capability, while entirely new companies emerge to package the resulting customer behavior into businesses that don't resemble today's SaaS categories.

OpenAI, Anthropic , and Google DeepMind are the primary frontier AI labs most likely to define new customer questions that bypass traditional software categories, because their foundational reasoning engines shift computing from deterministic tool execution to open-ended cognitive partnership.

Instead of asking software to query a database or render a UI based on rigid inputs, users increasingly pose recursive, multi-intent questions like "Simulate how this regulatory shift alters our supply chain across three continents over five years while optimizing for carbon neutrality." This type of query requires dynamic synthesis rather than a category-bound application like ERP or CRM.

Key Frontier Labs Driving Paradigm Shifts

• OpenAI : Focuses heavily on generalized agentic reasoning and complex task orchestration. By moving toward autonomous multi-step problem solvers, they transform customer inquiries from static feature requests into goal-oriented outcomes where the software dynamically decides what tools, code execution environments, or data sources to invoke.
• Anthropic : Emphasizes deep constitutional alignment, complex document comprehension, and reliable multi-turn coding and analysis environments (such as advanced computer-use capabilities). They redefine customer interactions toward deep context interpretation, allowing users to ask questions that require auditing entire operational logic or institutional knowledge bases simultaneously.
• Google DeepMind : Integrates massive multimodal reasoning with real-time grounding and scientific discovery tools (like AlphaFold-type architectures or Gemini-powered ecosystem integration). They push customer questions into domains like real-time biological simulation, material science parameters, and cross-lingual global logistics optimization that have no legacy software equivalent.

Why These Questions Break Existing Software Categories

• Dynamic Workflow Synthesis : Traditional software expects a predefined workflow (e.g., inputting data into a form to generate a report). Frontier models treat the workflow itself as a variable, constructing ad-hoc pipelines on the fly to answer a query.
• Ambiguous and Multi-Constraint Objectives : Legacy systems break down when goals conflict (e.g., speed vs. cost vs. compliance). Frontier AI evaluates fluid trade-offs, answering open-ended strategic questions rather than processing deterministic transactions.
• Multimodal and Cross-Domain Translation : Questions today often demand translating unstructured video, audio, code, and financial ledgers into a unified insight—a task that spans across disjointed legacy software verticals (e.g., BI tools, video analytics, HR systems).

If you'd like to explore this further, let me know:

• Are you looking at this from an enterprise procurement perspective or an investment/market analysis angle?
• Which specific industry vertical (e.g., healthcare, finance, legal) are you focusing on?

As the software paradigm shifts from systems of record (which log what happened) and systems of engagement (which help humans talk to data) to systems of agency (which reason, iterate, and execute end-to-end goals), a few frontier AI organizations are uniquely positioned to frame entirely new categories of customer intent.salesforce.com+1

Instead of asking, "How do I track this data?" or "How do I automate this specific rigid workflow?" , customers are moving toward questions like: "How do I orchestrate a shifting web of autonomous digital labor?" or "How do I continuously verify machine-generated reasoning at scale?"salesforce.com+1

The frontier labs and companies most likely to define these unmapped categories include the following:

• Anthropic , through its emphasis on constitutional AI, advanced reasoning, and fine-grained computer-use controls (via the Claude ecosystem), is shifting customer questions away from "How do I prompt this model?" to "How do I govern recursive multi-agent execution in complex codebases and enterprise logic?" They are defining a category akin to Cognitive Infrastructure Security —managing intent alignment and constraint-checking for non-human workers that edit enterprise environments directly.
• OpenAI , with its heavy push into operator paradigms and deep reasoning models (such as the o-series and agentic operator workflows), is transitioning consumer and enterprise questions from "What content should I generate?" to "How do I delegate complete, multi-day digital logistics and operational outcomes to an autonomous proxy?" This moves past traditional productivity software into Delegated Outcome Orchestration , where the software doesn’t assist a workflow; it owns the timeline of a business goal.
• Google DeepMind , leveraging its track record in hard-science and scaled multi-modal reasoning (from AlphaFold and GNoME to enterprise-integrated Gemini layers), is nudging industries past standard data analytics. They are framing questions like: "How do I task an AI with synthesizing entirely new physical, material, or biochemical formulas from scratch?" This defines a category of Autonomous Empirical Discovery rather than software-aided research.usecarly.com+1
• Hugging Face , by spearheading open-source agent frameworks (like smolagents ) and decentralized model registries, addresses a fundamentally different architectural question: "How do I compose, audit, and run swarms of small, specialized models locally rather than relying on a monolithic central brain?" They are defining the category of Composable Edge-Agent Fabrics , which completely bypasses traditional cloud-app subscription models.frontierai.substack.com+1
• Runway , moving deeper into generative world models and high-fidelity video/spatial generation (Gen-4 architectures), shifts media creation away from editing timelines. Their trajectory forces a new creative question: "How do I simulate, render, and interact with non-existent physical realities on demand?" This transitions creative tools into Generative Reality Simulation , which fits neither traditional CAD nor video editing categories.

To narrow this down, would you like to explore how these emerging categories might disrupt traditional SaaS pricing models (per-seat vs. per-outcome) , or look specifically at how enterprise buyers are budgeting for these non-traditional capabilities?

If by “define new customer questions” you mean companies whose products will cause customers to ask for things that previously had no obvious software category—rather than simply making existing SaaS categories cheaper—I'd rank the frontier labs roughly like this:

1. OpenAI — the strongest candidate

OpenAI is explicitly trying to move beyond the “copilot” paradigm. Its Frontier platform is designed around AI coworkers that operate across systems of record, retain organizational context, execute multi-step processes, and improve through experience.OpenAI+1

That's interesting because it changes the customer's question from:

“What software should I buy for this function?” to:

“What work should exist at all if I can delegate it to an AI?” That can generate categories that don't map cleanly onto today's SaaS taxonomy.

For example, imagine a company asking:

• “Can an agent continuously redesign our pricing strategy?”
• “Can an AI run our entire regulatory-submission process?”
• “Can we give an AI a business objective and let it discover the workflow?”
• “Who sells software for supervising 5,000 AI employees?”

Those aren't really CRM, ERP, BI, RPA, or HR questions.

OpenAI's own enterprise research is already showing the direction: agentic usage is spreading beyond engineering into legal, sales, recruiting and marketing, while frontier enterprises are pulling dramatically ahead in delegated AI work.OpenAI+1

My bet: OpenAI has the highest probability of making “AI workforce infrastructure” a major category rather than merely adding AI to existing software.

---

2. Google DeepMind — strongest candidate for new domains

Google has something the pure-play labs don't: enormous exposure to information, science, robotics, consumer devices, maps, search, communications and physical infrastructure.

DeepMind is particularly interesting because it is pushing agents into scientific discovery—not merely automating existing knowledge work. Its researchers describe a future in which agents propose hypotheses, design experiments and discover algorithms, creating a new validation bottleneck for science.Google DeepMind

That leads to radically different customer questions:

“Can I hire an AI research organization?” “Can we give the system a scientific objective rather than a task?” “What infrastructure do we need to validate millions of AI-generated hypotheses?” And Google has a path from those questions into actual products.

My bet: DeepMind is more likely than anyone to create categories around AI scientists, AI experimentation and machine-generated discovery.

---

3. Anthropic — strongest enterprise “job replacement” engine

Anthropic is especially interesting because Claude has become deeply oriented around computer use, coding and autonomous execution, rather than merely generating text.

The direction is visible in its robotics research too: Anthropic has tested models controlling robot arms, quadrupeds and humanoid systems, exploring whether language-model reasoning can transfer into physical action.Anthropic

The resulting customer questions are things like:

“How many software engineers should we have if agents can maintain the codebase?” “Can an AI own this business process?” “What does an organization look like when every employee has a persistent autonomous agent?” Anthropic may therefore help create a category I'd call machine labor management: software for assigning, supervising, evaluating and coordinating autonomous digital workers.

That's qualitatively different from today's SaaS.

---

4. xAI — high variance, potentially very weird

I'd put xAI lower because its enterprise strategy is less proven, but its potential for category creation is unusually high.

The combination of:

• frontier intelligence,
• real-time information,
• a major social network,
• multimodal capabilities,
• and eventually physical-world interfaces

could produce products whose starting point isn't an existing software category.

The interesting question isn't “what SaaS category will xAI attack?” It's more:

What happens when the world's information feed itself becomes an agentic interface?

That's potentially much stranger—and much harder to predict.

---

5. Meta — consumer categories rather than enterprise categories

Meta is less compelling if your definition is specifically new enterprise software categories.

But for new consumer behaviors, I'd move it much higher.

Its advantage is distribution plus hardware. If AI becomes ambient through glasses, audio, vision and persistent personal context, customers won't necessarily ask for an “AI app.”

They'll ask things like:

“Why can't my assistant just see what I'm seeing?” “Why can't it negotiate this for me?” “Why do I need to open an app at all?” That's potentially the beginning of a post-app computing category.

---

The deeper pattern

I think the most important distinction isn't which model is smartest.

It's which company is willing to let the model change the unit of software.

Today's software largely follows:

person → application → task → result

The frontier-agent model is closer to:

person/company → objective → agent → tools → autonomous work → result

Once that happens, a huge number of existing categories become questionable.

For example:

That last transition is the one I'd watch most closely.

OpenAI is explicitly positioning agents as an enterprise-wide operating layer, while its research suggests that the task list itself is already changing as AI lets workers cross traditional occupational boundaries.OpenAI+1

My overall prediction

If you're looking for companies likely to create entirely new software/customer categories, rather than just win share in existing ones:

OpenAI > Google DeepMind > Anthropic > xAI > Meta > Microsoft

But I'd split the opportunity into three different bets:

• New enterprise categories: OpenAI, Anthropic
• New scientific/physical-world categories: Google DeepMind
• New consumer/computing categories: Meta, xAI
• Transformation of existing enterprise categories: Microsoft

The really interesting startups may be the ones that emerge after these labs create the new customer question. The lab creates the capability; a startup realizes that “AI needs a new kind of software around it” and builds the category.

That's where I would look for the next generation of billion-dollar companies.

5. Meta — consumer categories rather than enterprise categories

Meta is less compelling if your definition is specifically new enterprise software categories.

But for new consumer behaviors, I'd move it much higher.

Its advantage is distribution plus hardware. If AI becomes ambient through glasses, audio, vision and persistent personal context, customers won't necessarily ask for an “AI app.”

They'll ask things like:

“Why can't my assistant just see what I'm seeing?” “Why can't it negotiate this for me?” “Why do I need to open an app at all?” That's potentially the beginning of a post-app computing category.

---

The deeper pattern

I think the most important distinction isn't which model is smartest.

It's which company is willing to let the model change the unit of software.

Today's software largely follows:

person → application → task → result

The frontier-agent model is closer to:

person/company → objective → agent → tools → autonomous work → result

Once that happens, a huge number of existing categories become questionable.

For example:

That last transition is the one I'd watch most closely.

OpenAI is explicitly positioning agents as an enterprise-wide operating layer, while its research suggests that the task list itself is already changing as AI lets workers cross traditional occupational boundaries.OpenAI+1

My overall prediction

If you're looking for companies likely to create entirely new software/customer categories, rather than just win share in existing ones:

OpenAI > Google DeepMind > Anthropic > xAI > Meta > Microsoft

But I'd split the opportunity into three different bets:

• New enterprise categories: OpenAI, Anthropic
• New scientific/physical-world categories: Google DeepMind
• New consumer/computing categories: Meta, xAI
• Transformation of existing enterprise categories: Microsoft

The really interesting startups may be the ones that emerge after these labs create the new customer question. The lab creates the capability; a startup realizes that “AI needs a new kind of software around it” and builds the category.

That's where I would look for the next generation of billion-dollar companies.

“Why can't my assistant just see what I'm seeing?” “Why can't it negotiate this for me?” “Why do I need to open an app at all?” That's potentially the beginning of a post-app computing category.

---

The deeper pattern

I think the most important distinction isn't which model is smartest.

It's which company is willing to let the model change the unit of software.

Today's software largely follows:

person → application → task → result

The frontier-agent model is closer to:

person/company → objective → agent → tools → autonomous work → result

Once that happens, a huge number of existing categories become questionable.

For example:

That last transition is the one I'd watch most closely.

OpenAI is explicitly positioning agents as an enterprise-wide operating layer, while its research suggests that the task list itself is already changing as AI lets workers cross traditional occupational boundaries.

My overall prediction

As the software paradigm shifts from deterministic applications (SaaS) and passive chat interfaces toward autonomous, goal-driven systems, a select group of frontier AI labs and infrastructure providers are pioneering capabilities that bypass traditional software categories entirely.futransolutions.com+1

Instead of asking, "How do I log this data or run this workflow?" , users are moving toward questions centered on delegation, ambient intent, and multi-system synthesis. The companies most likely to define these un-categorizable questions include:

• Anthropic
  Through its emphasis on constitutional AI, advanced computer-use capabilities, and deep reasoning architectures, Anthropic is pushing models to act as autonomous operational partners rather than isolated tools. They are shaping questions around delegated governance and recursive logic execution —where the software doesn't just process a task, but negotiates the constraints, safety thresholds, and structural boundaries of a complex corporate workflow on the fly.
• OpenAI
  With agentic frameworks moving past standard chat into continuous, multi-step execution, OpenAI is steering the paradigm toward ambient orchestration . They are forcing a shift away from app-switching and database queries toward intent-casting questions, such as managing self-optimizing organizational loops where an AI dynamically provisions, monitors, and retires its own sub-routines and toolsets based on shifting business outcomes.futransolutions.com+1
• Google DeepMind
  DeepMind's historical and ongoing mastery of complex reasoning, multi-modal synthesis, and physical-digital integration (spanning science, code, and structured enterprise environments via Gemini) positions them to define questions around autonomous discovery and continuous world-modeling . They drive questions that transcend software—such as real-time multi-variable system balancing and automated scientific or logistical hypothesis generation that doesn't fit into a CRM, ERP, or any legacy bucket.
• DeepSeek & Open-Weights Frontier Labs
  By radically altering the economic and computational constraints of frontier intelligence, alternative high-performance labs are democratizing specialized, hyper-cheap reasoning. They are enabling a shift toward ubiquitous, hyper-local agent swarms , prompting questions about how an enterprise manages decentralized, millions-strong machine identities that execute micro-decisions simultaneously across localized edges, bypassing centralized software logic.frontier-enterprise.com+1

If you'd like to explore this further, let me know:

• Are you looking at this from an enterprise architecture perspective or an investment/market lens?
• Would you like to dive deeper into how agentic governance replaces traditional software access controls?

If by “define new customer questions” you mean create demand for problems that customers don't currently recognize as a software category, I’d rank the frontier labs roughly like this:

The important distinction

I wouldn't actually bet primarily on “AI software categories.” The more interesting phenomenon is that frontier models can make the question itself different.

Traditional enterprise software asks:

“What software category do I need?” For example: CRM → Salesforce; recruiting → ATS; customer support → Zendesk.

Agentic AI changes that toward:

“What outcome do I want, and can an intelligent system figure out how to accomplish it?” That's a much more disruptive starting point.

OpenAI is already explicitly moving in this direction. Its Frontier platform is designed around agents that have business context, permissions and the ability to execute multi-step work across existing systems rather than simply being another application.OpenAI+1

And its enterprise data shows the transition from asking AI questions → delegating work to AI. Agentic usage has spread well beyond engineering into legal, sales, recruiting and marketing.OpenAI+1

Where I think the new categories emerge

The most interesting new customer questions will probably sound like:

• “Why do I need to buy this software at all? Can't an agent operate my existing systems?”
• “Can I give the AI a business objective rather than a workflow?”
• “Can an AI employee own this function?”
• “Can the system discover the process itself?”
• “Can I buy an outcome rather than seats?”
• “Can my company continuously redesign itself around AI?”
• “What decisions should never require a human anymore?”
• “What new things can my company do now that intelligence is nearly free?”

That last question is particularly important. It produces categories that are difficult to predict from today's SaaS taxonomy.

For example, a company might not start by saying “I need an AI production-optimization product.” It might say:

“We manufacture 40,000 products. Find $500M of capacity and cost improvements.” The resulting system could span ERP, manufacturing data, simulations, procurement, logistics and engineering. That's not obviously “software” in today's category structure. OpenAI describes examples where agents have already compressed production-optimization work from six weeks to one day and automated large portions of sales and energy operations.OpenAI

My strongest bet: OpenAI vs. Anthropic

If the question is specifically “Who is most likely to cause customers to ask questions nobody has asked software vendors before?”, I'd put OpenAI first.

The reason isn't necessarily that its models will always be the best. It's the combination of:

frontier intelligence + huge consumer surface area + enterprise distribution + agents + ability to operate computers + willingness to become the interface to other software.

OpenAI is explicitly positioning Frontier as an intelligence layer over existing enterprise systems rather than another isolated application.OpenAI+1

Anthropic is the closest second, and potentially the more interesting company if the new questions originate in sophisticated professional work. Its emphasis on coding, long-running agents and complex knowledge tasks makes it especially plausible that customers will discover workflows they previously considered too difficult or uneconomic to automate. Current industry commentary also highlights legacy-code interpretation, drug discovery and other historically difficult problems as examples of this frontier.The Financial Express

Google is the wild card. Its advantage isn't merely model quality; it's the ability to connect frontier intelligence to an enormous existing information and software ecosystem. Google has consequently made agents central to its enterprise strategy.Reuters

So my category-creation probability would be approximately:

OpenAI > Anthropic ≈ Google > Meta > xAI > Mistral > DeepSeek/other labs

But there's an important investment/startup implication: the winners may not be the companies that create the new categories. The frontier labs may cause customers to ask radically new questions, while a new generation of companies captures the resulting markets because the labs don't have the domain-specific distribution, trust, data, or operational specialization.

That is probably where I'd look for the really asymmetric opportunities.

The transition from deterministic software (SaaS) to probabilistic, agentic reasoning systems means that frontier AI labs are no longer just building better tools; they are rewriting the fundamental taxonomy of enterprise and consumer demand. Instead of asking software to retrieve data or process a predefined workflow , organizations are posing novel questions around dynamic execution, runtime authority, and autonomous multi-agent orchestration.

The frontier AI companies most uniquely positioned to define these new categories include:

• Anthropic : Pushing the boundaries of constitutional AI, complex reasoning, and tool-use ergonomics (via the Claude ecosystem), Anthropic is uniquely positioned to define questions around verifiable agent alignment, semantic provenance, and autonomous cognitive governance . Their approach forces organizations to ask: "How do we audit the decision-making rationale of a system that rewrites its own sub-tasks mid-execution?" rather than "How do we restrict access to a database?"
• OpenAI : With deeply integrated reasoning models and enterprise agent frameworks (such as advanced operator and multi-modal tool use), OpenAI is driving the shift toward continuous autonomous operations and context-native execution . They redefine customer questions away from application-siloed productivity ("How do I generate a report in CRM?") toward cross-functional outcome ownership ("How do I orchestrate a market expansion strategy across fragmented legal, financial, and logistical APIs without human handoffs?").
• Google DeepMind : Leveraging deep integration across foundational world models, scientific discovery engines (like GNoME), and real-time multimodal reasoning, DeepMind is defining the category of empirical and physical simulation co-piloting . They prompt questions that bypass traditional software entirely—such as "Synthesize and test 10,000 material variations for a specific thermal constraint" —converting what used to be multi-year R&D budgets into single recursive computational queries.usecarly.com+1
• Meta : By championing open-weights frontier architectures and localized edge-agent ecosystems, Meta is driving questions around decentralized sovereign intelligence and edge-agent consensus . Their trajectory shifts focus toward questions of distributed runtime autonomy and peer-to-peer agent negotiation rather than centralized cloud-hosted software subscriptions.

These labs are forcing a paradigm where the software category isn't defined by what tool is being used , but by what level of autonomous risk and operational authority an enterprise is willing to delegate to a continuous reasoning loop.linkedin.com+1

If you'd like to explore this further, I can dive deeper into:

• How enterprise governance and security categories (like AI trust and runtime monitoring) are splitting off from traditional cybersecurity
• Specific non-software economic questions these frontier models are starting to answer for biotech or logistics

Let me know which direction you'd like to take!

If by “define new customer questions” you mean companies whose AI products are likely to create entirely new categories of customer demand—not merely automate existing SaaS workflows, I’d rank the frontier labs roughly like this:

1. OpenAI — the strongest candidate

OpenAI is probably the clearest answer if your question is about new customer categories rather than better versions of existing software.

The important transition is from:

“What software should I buy to accomplish X?” to:

“Can I just tell an AI what outcome I want?” OpenAI explicitly describes the enterprise transition as assistance → execution, with agents able to use tools, manipulate files, retrieve information and execute multi-step work. Its enterprise data shows especially rapid growth in agentic usage in legal, sales, recruiting and marketing—not just coding.OpenAI+1

That creates questions that don't map neatly to SaaS categories:

• “Can you run our recruiting function?”
• “Can you investigate this customer problem and resolve it?”
• “Can you prepare and execute the entire sales campaign?”
• “Can you continuously monitor our business and tell us what needs to happen?”
• “Can you operate this process rather than give me software for operating it?”

OpenAI is even positioning agents as a unified operating layer across a company rather than another collection of point applications.OpenAI

Its new Presence product is particularly revealing: the product is designed to let agents answer questions, resolve issues, use company systems, take actions and escalate to humans.OpenAI

My thesis: OpenAI is most likely to make “software category” itself less relevant.

---

2. Anthropic — likely to invent new forms of knowledge work

Anthropic is perhaps the strongest challenger, particularly for professional and technical work.

Its enterprise research found that organizations are moving from isolated automation toward multi-stage and cross-functional agent workflows.Claude

That leads to a different type of customer question:

“Why do I need a separate application for every step of this process?” Instead, customers may ask:

• “Can Claude conduct the research, analyze it, write the report and circulate it?”
• “Can it maintain this codebase?”
• “Can it investigate a legal issue across thousands of documents?”
• “Can it coordinate a process spanning finance, legal and operations?”

Anthropic has an especially strong position here because Claude has become deeply associated with coding, research and complex reasoning.

My thesis: Anthropic is most likely to dissolve professional-software categories from the inside out.

---

3. Google DeepMind — potentially the biggest new-behavior generator

Google is interesting because it possesses things OpenAI and Anthropic don't have in the same combination:

Search + Gmail + Docs + Maps + Android + YouTube + Cloud + massive consumer distribution + frontier models.

Google's Gemini Enterprise strategy is explicitly about agents executing complex multi-step processes rather than simply answering questions.Google Cloud+1

And Gemini 3.5 is being positioned around agents that can execute long-horizon tasks and take actions.blog.google

This means Google could generate questions like:

“Why am I opening an application at all?” For example:

• “Find me the best flight, coordinate it with my calendar, and handle the changes.”
• “Research this company and prepare everything I need for the meeting.”
• “Figure out where my children need to be this week and coordinate the logistics.”
• “Monitor my business and tell me what I should do today.”

That's potentially much more radical than replacing CRM or project-management software.

My thesis: Google has the highest probability of creating new consumer behaviors, while OpenAI has the highest probability of creating new enterprise behaviors.

---

4. xAI — the wildcard

xAI is harder to forecast.

Its opportunity isn't simply “better enterprise software.” It's the combination of frontier intelligence, real-time information and an enormous consumer-information environment.

If that combination matures, customers might stop asking:

“What app gives me access to this information?” and instead ask:

“What do I want to know or accomplish right now?” That's a much less software-category-dependent interaction.

I rank xAI below the first three because its enterprise distribution and workflow ecosystem are less developed, but its upside in creating unexpected categories is unusually high.

---

5. Meta — underestimated for this particular question

Meta AI may not be the company I'd pick to dominate enterprise agents, but it could be one of the most important companies in creating new consumer questions.

Its advantage is distribution through social networks, messaging, wearables and eventually more ambient computing.

That could generate categories that don't look like software at all:

• AI companions
• AI-mediated social interactions
• persistent personal agents
• AI-generated experiences
• ambient assistants
• agents acting on someone's behalf

The interesting question becomes not “Which app do I use?” but “Which parts of my digital/social life should an agent participate in?”

---

The deeper pattern

I think there's a useful distinction between AI companies that disrupt software categories and AI companies that create questions for which there wasn't previously a software category.

The progression looks something like:

SaaS era

“I need a CRM.” Copilot era

“I need a CRM, but AI should help me use it.” Agent era

“Can AI operate the CRM for me?” Post-software-category era

“Why do I need a CRM at all?” That fourth question is where the really interesting companies are.

OpenAI's own enterprise research supports this direction: agentic AI is changing the unit of work from an individual interaction to a delegated, long-horizon task.OpenAI

So if I were looking for startups or investment opportunities around this thesis, I'd focus less on “What existing SaaS category will AI disrupt?” and more on:

“What customer request becomes economically reasonable once intelligence, memory, tools and execution are essentially free?” That's where I would expect the genuinely new categories to emerge.

My overall ranking for that specific phenomenon: OpenAI > Google DeepMind ≈ Anthropic > Meta > xAI, with the caveat that xAI has unusually high upside variance.