This video examines what companies will keep buying when AI agents perform work, and how existing SaaS apps may survive or weaken. Analyze software through data, application and workflow, and agent layers, rather than treating it as simply an app. Employees, buyers, and vendors must all explain the outcomes they produce with AI. 🤖

1. DoorDash as a Glimpse of Software's Future

The presenter unexpectedly starts with DoorDash as a key to software buying and selling in 2026–2027. Ordering food by message is more than convenience: it demonstrates services working without users opening the app.

"You can message DoorDash and have it bring what you want, your favorite pizza or sandwich. This is software's future."

At work, an agent could read a new customer's contract, sales notes, and email threads to prepare onboarding, instead of employees opening several tools. Information processing moves to agents and humans need fewer app screens.

Companies holding customer records and core data still have value and can charge. But extra apps and interfaces that merely display data or perform repetition become easier to replace.

"The extra tool, the software you used to make that plan, becomes less important and easier to leave. That's the risk facing SaaS."

Context employees teach agents becomes a powerful asset: one customer requires security approval before onboarding; one team works outside standard procedures. Such undocumented knowledge makes trained agents harder to replace.

"The agent you've trained may be much harder to replace because lessons are stored in your personal system or work agent's memory."

The conflict is therefore not merely which model is cheaper. A finance or IT decision to switch providers may also destroy established workflows and productivity.

2. How a Recruiting Agent Becomes Hard to Replace

The presenter offers a partly anonymized recruiting example. A recruiter traditionally searches a tool weekly, copies notes into a spreadsheet, and makes a hiring-manager shortlist. Agents change that flow.

The agent retrieves profiles from a candidate-data provider and prepares a list for human review and approval. The provider still earns revenue from data, but the recruiter increasingly stops opening the old tool.

"The company still sells candidate data and you still use it. But you begin paying for different units of work performed on top of that data."

The bigger change comes as the agent learns judgment: cultural fit, preferred experience combinations, individual managers' preferences, and who has already been hired. It becomes a colleague tailored to personal recruiting philosophy and organizational context, not merely search.

"It learns how you judge successful candidates, rule out poor fits, and define an ideal candidate."

This extends to product, engineering, sales, and almost every field. Formerly single-purpose tools become easier to leave, while long-trained agents become harder to abandon.

Human review remains necessary. Responsible, ethical users must still check recommendations themselves. Reduced work does not transfer final judgment and responsibility entirely to AI.

3. Why People Still Pay in the Agent Era

The presenter asks when viewers actually open an app themselves, rather than instruct an agent: when humans directly find, organize, and act on information.

"When do you, as a human, find information, organize it, and act on it?"

SaaS vendors often bundled search, organization, action, and storage as value. Once agents handle even part of that, customers ask what exactly they were paying for. AI clarifies the substance of existing value.

Buyers should first examine vendors' data and distribution capabilities. "We have AI features" is a baseline. What position can the vendor hold using its existing data, customer access, and trust?

"To understand what's happening as a buyer, start with the seller's data and distribution."

Purchasing in 2026 becomes more complex: assess both today's product and how well it will connect and scale as agents improve. A poor choice may disrupt productive workflows employees already built.

4. Meta and Salesforce's Different Strategies

Meta, though not a traditional SaaS company, is entering work software by rapidly extending personal agent Muse into small businesses.

Connections to Shopify and QuickBooks give Muse commerce and accounting data. Meta aims beyond personal assistance toward becoming the operational headquarters for finding customers and doing work.

"Meta seems to want to expand from personal assistant into finding customers and helping run operations."

Personal and workplace AI boundaries are disappearing quickly. Employees bringing personal agents to work can move consumer productivity into enterprise software rapidly.

Salesforce starts with longstanding CRM data and strong enterprise distribution, strategic assets that remain valuable.

It introduced Koa, a reasoning model tuned for CRM. Koa trains on synthetic, not actual customer data, but Salesforce's long experience with data and workflows informs its understanding of how real customer information works.

"Better models let Salesforce apply knowledge already in its systems to more customer work."

It seeks all three layers:

• Data: Core customer-relationship and sales information
• Applications: Established CRM products and workflows
• Agents: An environment managing and connecting Koa and other agents

The presenter sees Salesforce enabling Claude, Cohere, and other models instead of forcing one. Choice and multiple models appeal to enterprises avoiding lock-in.

"Buyers don't want to be locked in. Sellers therefore need to sell options."

5. Microsoft and the AI Labs

Microsoft faces a similar dilemma, owning Azure infrastructure, development tools, Outlook, Excel, Teams, Copilot, and Autopilot.

As people ask agents to process data instead of opening Excel or Outlook, screen-centered products must redefine themselves. Microsoft offers a trusted integrated stack, not an isolated product.

"We manage your data, offer safe storage in Azure, provide familiar productivity tools, and put trusted AI on top."

The message is that Microsoft already has customers' data, development, productivity, security, and trust, and now offers future work through Copilot and Autopilot. Large enterprises' concern for stability and accountability makes existing trust especially powerful.

OpenAI and Anthropic start elsewhere. They have neither longstanding enterprise work data nor equivalent established enterprise distribution. Their proposition is better intelligence and applying it to customer data.

"The labs' plan is simple: bet on our intelligence and our ability to apply it to your data."

Anthropic has focused more strongly on enterprises; OpenAI is expanding from consumer success. Salesforce-Claude cooperation is natural: Salesforce brings data and customer access, while Anthropic brings models and field engineering.

6. DoorDash and MCP: Winning Transactions Without App Opens

DoorDash wants people to use its assistant, but more importantly wants orders even through other companies' agents. It is testing MCP, Model Context Protocol, alongside text ordering.

MCP connects agents to outside data and tools. DoorDash cares more about receiving the order than whether someone opens its app.

"DoorDash wins even if you never touch its app again. Losing the interface moment doesn't mean losing the transaction or customer value."

A user might type /DoorDash in a command or AI window to order a usual lunch, or order by text. DoorDash loses app time but keeps transactions.

Not every company wins merely by offering an MCP server. Agent connectivity is a minimum requirement for consideration, not a reason to keep paying. Vendors need proprietary data, reliable execution, complex workflows, or brand trust.

"Being connectable to agents only gets you considered. It doesn't explain why someone should keep paying."

7. Employees' AI Habits Change Buying Decisions

One practical message is not to hide personal AI workflows. Employees combine tools, prompts, automations, and data access into individual productivity systems.

Keeping them as secret weapons creates risk. A company may change vendors for cost or security without knowing how dependent employees are on specific tools and agents.

"Don't hide it. If the provider changes, you're the one who loses."

Switching based only on token or subscription costs may destroy months of accumulated context and automation. This does not mean everyone should report directly to the CFO. Tell managers so they can represent real requirements in higher-level decisions.

No elaborate technical explanation is needed. Explain which data and apps you use, where agents save time, and which workflows matter.

"Talk about how you use AI: the data, workflows, when you actually click into apps and when you don't."

8. Crustdata and Workday: Two Kinds of Value

A two-person recruiting firm using Crustdata builds its own AI workflow through an API rather than buying a complete recruiting tool.

It supplies selection criteria to AI; Crustdata supplies structured career and candidate information. The firm pays for data and its structuring, while building search and screening workflows itself.

"They chose to keep buying structured, agent-readable data and build their own research workflow on top."

AI has lowered custom-software costs. Small firms once found ready-made SaaS easier; good data now makes tailored workflows feasible.

Workday payroll is different. Explaining a payslip is not the same as reliably paying employees in compliance with rules. Payroll includes regulations, records, approvals, execution, and exceptions.

"AI explaining a payslip is entirely different from replacing the system ensuring people get paid."

Workday's payroll agent can detect problems and suggest corrections using enterprise data and rules. But an established record, approval, and execution system underpins it. Regulated, high-risk workflows are trusted operational systems on data, not mere screens, and will not disappear easily.

"Agents may make these systems more valuable by helping people use them better, rather than less valuable."

9. The Three Layers: Data, Applications, and Agents

The video's central framework is:

1. Data
   Customer information, candidate profiles, transactions, and payroll rules are work's raw material. Agents need trustworthy, accessible data.

2. Applications and workflows
   Rules, approvals, procedures, and execution systems do real work correctly. Screens may change, but complex, regulated workflows retain value.

3. Agents
   Agents read data, invoke workflows, act on context, and accelerate work. How individuals and teams train and use them increasingly matters.

"We all need to explain our work in terms of data, application, and agent layers."

Employees need to understand tools and important data. Buyers must recognize different organizations' data, applications, and team practices. Sellers must explain their value across the layers, beyond claiming AI features.

Applications do not simply end. The threatened experience is interface-centered work requiring people to open screens and repeatedly click. Reliable workflows on data remain valuable.

"Separate applications into interfaces and workflows. Workflows operating on data can be extremely valuable."

Nor will everyone use command lines. Many knowledge workers are unfamiliar or uncomfortable with them. For now, AI will integrate into familiar interfaces, and users will mix screens with agent workflows.

10. What Employees, Leaders, and Vendors Should Do

Employees should explain where AI works, which data matters, when they must open apps, and what agents can handle. Sharing helps both the team and the company's ability to support them.

"Sharing what works for you improves the team and helps the company support you better over time."

Leaders and buyers need a deep understanding of the team's actual "AI footprint." Board-level declarations of adoption are insufficient. Standardization or cost-cutting can destroy productivity if actual workflows are overlooked.

"You can't sit at board level saying, 'We understand AI; we'll put it on top of everything.' That doesn't work."

Vendors and builders must sell more than today's features. Customers buy confidence that the product remains useful when the next model arrives. Explain access across interfaces and agents, workflow compatibility, and cooperation with other agents.

"People buying today are buying belief that you'll keep up when tomorrow's next model arrives."

Companies ultimately buy faster outcomes: revenue, profit, customer experience, or development. Employees may find this uncomfortable, because organizations will ask how AI accelerates the whole team's work, not merely whether it is used.

"They're not looking for AI adoption. They're asking whether all of us can accelerate the whole operation."

11. Closing: Everyone Becomes an AI Buying Decision-Maker

Agents do not end software; they subject its value to stricter scrutiny. Data suppliers, complex-workflow providers, and personalized-agent builders can survive differently. Mere screens or vague convenience may no longer justify former prices.

Like the recruiter, individual employees become decision-makers. Their tools, connected data, and trained agents affect company-wide purchases and switching costs.

"Agents force us to think vertically across the whole chain. Everyone is now a decision-maker."

The essential questions are: Who supplies data? Who supplies trusted workflows on it? Who supplies the agent layer? Clear answers help individuals and companies make better choices and create stronger value in the agent-era software market.