Software for the AI Era

By Zvi Schreiber · April 2026 · 10 min read

Three Shifts Reshaping Business Software — and Why They’re Connected

Something is broken in how companies adopt AI agents.

Consider a growing services business, about 30 people that recently deployed AI across their operations. They added an AI assistant to their CRM. They connected a chatbot to their support tool. They integrated an AI writing tool for their marketing team. Each tool worked ok in isolation. The CRM assistant helped salespeople draft better emails. The support chatbot resolved simple tickets. The marketing AI produced decent first drafts.

But the support chatbot has no idea what the sales team had promised the customer. The CRM assistant doesn’t know about the open support tickets. The marketing AI is writing about product capabilities that the engineering team had quietly deprioritized. Each AI is competent within its silo and blind to everything outside it.

They’ve built a faster spider — AI-enhanced legs, same fragmented body.

This is what happens when you bolt AI onto the existing software landscape without questioning three assumptions that have shaped business software for the past two decades. All three are now breaking simultaneously, and the organizations that recognize this will have an enormous advantage over those that don’t.

Assumption One: Your Tools Should Be Specialized and Separate

The best-of-breed philosophy has dominated small and mid-size business software since the early 2000s. Pick the best CRM. Pick the best accounting tool. Pick the best tasks tool. Pick the best support desk software. String them together with Zapier and hope for the best.

This made some sense when each tool was genuinely better at its job than any integrated alternative. Salesforce was better at sales than any ERP’s CRM module. QuickBooks was better at accounting than any all-in-one suite. The trade-off — fragmented data, integration headaches, humans as the glue — was worth it because the individual tools were so much better.

But the costs of that trade-off have compounded quietly. The average small business now has dozens of SaaS subscriptions. Each has its own data model, its own user management, its own permission system, its own reporting. The employee who wants to get a full picture of a customer — their purchase history, their open support tickets, their outstanding invoices, their upcoming contract renewal — has to log into four different tools, cross-reference data mentally, and hold the unified picture in their head.

Our industry built software to reduce cognitive load. Then we deployed so much of it that navigating the software landscape became its own cognitive burden.

Integration platforms helped, but they solve symptoms, not the disease. Zapier can push a closed deal into your invoicing tool. It can’t reason about whether this customer’s payment terms should be adjusted based on their support history and market conditions. Data warehouses can produce unified dashboards. They can’t enable unified workflows. The data lives together in the warehouse but works separately in the tools.

Broader suites exist — Zoho, Microsoft SharePoint — and they reduce fragmentation somewhat. But most were built as collections of tools acquired or developed separately, then connected after the fact. The boundaries between modules often remain visible: different UIs, different data stores, different permission models stitched together with internal APIs.

None of this mattered enough to force a reckoning — until AI.

AI amplifies the cost of fragmentation exponentially. An AI agent that can only see your CRM data is like an employee who’s only allowed into one room of the office. It can be useful within that room, but it can never develop the cross-functional understanding that makes humans valuable. The sales AI doesn’t know about the support problem. The support AI doesn’t know about the renewal opportunity. The finance AI doesn’t know about either. And unlike a human, an AI agent cannot walk to the next room and chat to a colleague from the next department.

Assumption Two: Your Team Is All Human

Every SaaS application you use was designed with a single type of user in mind: a human being sitting at a computer, navigating a graphical interface, making decisions with a brain that retains context between sessions.

That assumption is no longer accurate.

The modern team includes AI agents — not as tools that humans invoke, but as autonomous actors that own work, access data, and produce deliverables. A support agent triages and responds to tickets. A recruiting assistant screens candidates and schedules interviews. An operations agent runs daily reports and flags anomalies. These aren’t hypothetical. They’re already doing the work in thousands of organizations and there’s a lot more of them coming.

But the software they’re working within wasn’t designed for AI agents. The results are predictably awkward.

Agents as afterthoughts. Most platforms treat AI as a sidebar feature — a copilot, an assistant, a “smart” button. The AI doesn’t have its own identity within the system. It doesn’t have its own permissions. It doesn’t have its own audit trail. It’s a feature of the human’s session, not a user in its own right. This means you can’t assign work to an AI agent the way you assign work to a human. You can’t review what the agent did independently. You can’t configure its access separately from the human who invoked it.

No shared workspace. When a human and an AI agent both work on the same problem, they often can’t see each other’s contributions in context. The human makes changes in the UI. The agent makes changes through an API/MCP. The two experiences are disconnected — different views of the same data, with no common ground for collaboration.

No governance model. When a new employee joins your company, you have a well-understood process: define their role, set their permissions, assign a manager, give them a probationary period, expand their authority as they prove themselves. When you deploy an AI agent? You give it an API key and a system prompt. There’s no equivalent of the employment lifecycle — no structured onboarding, no progressive autonomy, no performance review, no explicit scope of authority.

What does a platform designed for hybrid teams actually look like?

Agents as first-class users. An AI agent has its own account, its own identity, its own permissions, and its own audit trail — just like a human employee. When the agent creates a task, it appears on the team’s task board with the agent’s name on it. When the agent reads customer data, the access is logged just like a human’s access would be. The agent isn’t a script running in the background; it’s a visible, accountable member of the team.

Shared workspace. Humans and agents work in the same system on the same data. An agent creates a task — the human sees it on their board. A human closes a deal — the agent sees it and can act on it. A human writes a note on a customer record — the agent can read it in context. No sync layer, no webhook, no integration. The workspace is shared because there was never a separation to bridge. And of course, for agents like for humans, sharing is scoped and controlled securely.

Handoff and collaboration. Work flows naturally between humans and agents. An agent handles a support ticket, hits something it can’t resolve, and escalates to a human — with full context, not a summary. The human resolves the issue and hands the follow-up back to the agent. A human can “act as” an agent, seeing the system from the agent’s perspective — invaluable for debugging, quality review, and seamless takeover. This isn’t a feature; it’s a fundamental design principle.

Unified governance. Every agent is configured through a consistent framework — in our case, the OCTOPUS model — that covers Objectives, Context, Tool access, Oversight, Processing, User relationships, and Scheduling. The same governance model applies whether you have one agent or twenty. There’s no separate “AI management console” — agent configuration is part of the same system where everything else happens.

Security through architecture, not afterthought. Agents inherit the same permission system as humans. Relationship-based access control means an agent’s data access is controlled exactly like a human team member’s. Whether the actual scope is more or less is up to the human. A sales agent can see deals but not payroll. An HR agent can see employee data but not customer tickets. Progressive autonomy — from sandbox mode to trusted operation — is configurable per agent, per capability. Every action is audited. This isn’t a security feature bolted onto an AI tool; it’s the same security infrastructure that protects all data in the system, applied consistently to all users, human and AI alike.

Assumption Three: You Pay Per Seat for Software Access

Per-seat pricing was an elegant innovation for human-only teams. It aligned cost with usage: more employees meant more usage, which meant more value for the software customer and more revenue for the software vendor. Fair enough.

But per-seat pricing creates a perverse incentive in the age of AI: it penalizes automation. If your AI agent counts as a “seat,” deploying automation increases your software costs. If it doesn’t count as a seat, the vendor captures none of the value their platform enables. Either way, the model breaks.

An AI-native pricing model looks different. The platform itself — the CRM, the task management, the accounting, the support desk, all of it — is the infrastructure. It’s the table stakes. The value is in the intelligence layer: the AI agents that do real work, eliminate expensive manual tasks, and enable a small team to operate as a much larger one.

When an AI agent spends a fraction of a cent to classify and route a support ticket — replacing five minutes of human triage — the economics are obvious. When it spends a few cents to reconcile a batch of invoices — replacing an hour of manual work — the value is even clearer. Every AI action replaces manual work that costs 10 to 100 times more.

This shifts the pricing model from an “access tax” (you pay for the privilege of logging in) to “value delivered” (you pay when the system does useful work for you). It aligns incentives: the software vendor succeeds when AI does more valuable work for the customer, not when the customer hires more humans.

It also makes powerful software accessible to smaller companies. A 15-person business that could never afford enterprise-grade tools at $50-100 per seat per tool can now access a complete business suite for free, paying only for the AI work that saves them time and money. The software cost scales with the value delivered, not with headcount.

Why These Three Shifts Are One Story

Here’s the insight that ties everything together: these three shifts aren’t independent trends that happen to coincide. They’re causally connected. Each one enables and requires the others.

You can’t have useful AI agents without unified data. An agent that sees only one module’s data makes the same mistakes as a human who only talks to one department. Shift 1 (unified platform) is a prerequisite for shift 2 (hybrid teams). The agent needs one nervous system to think across the business, not forty APIs to query separately.

You can’t justify a free platform without AI revenue. The traditional SaaS business model charges for software access. If you give the software away, you need a different revenue source. AI agent usage — priced at the value it delivers — provides that source. Shift 2 (hybrid teams doing real AI work) enables shift 3 (value-based pricing that makes the platform free).

You can’t compete with free on per-seat pricing. Once a unified platform offers the same capabilities — CRM, tasks, accounting, support, recruiting — for free, the per-seat model faces existential pressure. Shift 3 (new economics) makes shift 1 (unified platform) viable for small companies who couldn’t afford enterprise suites.

The three shifts are a package deal. And that’s why the organizations that recognize this early will have an outsized advantage. They won’t just have better software — they’ll have a fundamentally different operating model.

What This Looks Like

Consider a growing professional services company — 25 people, doing well, drowning in operational overhead.

Today, they run Salesforce for sales, Zendesk for support, QuickBooks for accounting, Greenhouse for recruiting, Asana for project management, and Slack for communication. Six tools, six subscriptions, six permission systems, six data silos. The office manager spends hours each week copying data between systems. The CEO maintains a mental model of the business by synthesizing information from multiple dashboards. When a client calls with a problem, the support team has to check three different systems to understand the full picture.

In the AI-native model, all of this lives in one platform. But more importantly, the team includes AI agents. These agents aren’t using different tools than the humans. They’re in the same system, seeing the same cross-departmental data (with appropriate permissions), producing work the humans review in the same interface. The support agent’s ticket resolution appears in the same queue the human agents use. The operations agent’s morning report appears in the same workspace where the team plans their day. The recruiting assistant’s candidate screening appears in the same pipeline the hiring manager reviews.

The company pays nothing for the software platform. It pays for the AI work the agents do — pennies per action, replacing hours of manual work. The total AI cost is a fraction of what they used to spend on SaaS subscriptions, and every dollar of AI spend directly replaces human effort that costs far more.

The Transition

The good news is that this transition doesn’t require a big bang migration. A company can start with one or two modules — move their CRM and support into a unified platform, deploy one AI agent in sandbox mode, and expand from there. Each module they add reduces fragmentation. Each agent they deploy reduces manual work. The value compounds.

The harder part is the mental shift. For twenty years, we’ve been trained to think about software as a collection of specialized tools. “What’s the best CRM?” is the wrong question. The right question is: “What system will enable my team — humans and AI agents together — to work most effectively?”

In the first paper in this series, I argued that AI enables a new organizational model — the octopus — where unified intelligence replaces the hierarchical concentration (spider) or edge distribution (starfish) of scarce human attention. In the second paper, I described the OCTOPUS framework for configuring individual AI agents within that model.

This paper is about the body the octopus needs: a unified platform where data, workflows, permissions, and intelligence — human and artificial — are part of one system. Not a spider enhanced with AI legs. Not a starfish with AI nodes. An octopus: one mind, many arms, no skeleton, endlessly adaptable. And that is PanLuma.

The era of the human-only team running on fragmented software is ending. What comes next is more interesting.

Zvi Schreiber has spent two decades building technology companies and is now focused on building software for the age of human-AI collaboration.