Is AI the new SaaS? Why the future is a conversational layer on top of your work
The SaaS industry was built on the insight that workflows could be systematised in software.
The AI layer is built on the insight that workflows could be described in language; and that language, processed by AI, could operate across every SaaS tool simultaneously without requiring a separate interface for each one.
This does not make SaaS irrelevant. It makes the SaaS tools that cannot be queried by language increasingly irrelevant.
The average $15M company pays for 22–35 SaaS tools. Most of those tools solve a specific workflow problem by providing a structured interface to perform specific actions on structured data. AI does something different: it provides a conversational interface that can query, act on, and synthesise data across all of those tools simultaneously.
That is not a replacement. It is a layer.
What “SaaS” was: and what it was actually solving
SaaS tools solved two distinct problems.
Problem 1: Data structure and storage
Before CRMs, contact data lived in spreadsheets, email, and people’s heads. Before accounting software, financial data lived in Excel files. SaaS tools created structured, queryable repositories of operational data.
This problem remains solved by SaaS. The data still needs to live somewhere structured. CRM data lives in the CRM. Financial data lives in the accounting tool. AI does not replace the need for structured data storage.
Problem 2: Workflow interface
The SaaS interface; the dashboards, the reports, the filter views, the click-through workflows; was how users interacted with and acted on the structured data. The HubSpot pipeline view. The QuickBooks invoice dashboard. The Asana board.
These interfaces let non-technical users perform structured operations on structured data.
This is the problem AI is changing.
A user can now say “show me the deals that have been stalled for more than two weeks” instead of building a filter in HubSpot. They can say “what is our cash position versus last month?” instead of navigating to the QuickBooks report.
The conversational layer replaces the click-through interface; but not the underlying data.
The net effect:
- SaaS tools that are primarily data storage and structure (systems of record) become more valuable; they are the data layer the AI queries
- SaaS tools that are primarily dashboards, reports, and interface layers on top of data that the AI can access directly become less valuable; the AI provides the interface
The three tool categories: which SaaS is getting stronger, which is getting weaker
Category 1: Systems of record (becoming more essential)
These are the tools where the operational data lives. They become more important as AI layers demand better, more accessible data.
Examples: CRM (HubSpot, Salesforce, Close), accounting (QuickBooks, Xero), project management (Monday, Asana), HR and payroll systems, email and calendar platforms.
The key attribute: they have APIs that AI tools can connect to. The CRM that does not have an API is a liability in an AI-first stack; the data is trapped. The CRM with a strong API is increasingly the foundation of the AI-assisted sales workflow.
What to look for when evaluating these tools: API accessibility, data export flexibility, and whether the AI tools the company is building are already integrating with them natively.
Category 2: Interface and reporting layers (at risk of replacement)
These are tools whose primary value is the interface or reporting view they provide over data that could be accessed another way.
Examples: standalone BI and reporting tools (Tableau, Looker for small-scale deployments), standalone dashboard tools, CRM reporting add-ons, analytics overlays that sit on top of the same data the CRM already holds.
The at-risk signal: if the primary value of the tool is “it shows us our data in a nice way” and the AI layer can produce the same view conversationally, the tool is at risk.
Ask: “if AI could answer any question we currently look at this dashboard to answer, would we still need this tool?”
Category 3: AI-native tools (growing, replacing category 2)
These are tools built around AI as the core interface; where the interaction is conversational or AI-driven from the ground up, rather than retrofitted onto a traditional SaaS interface.
Examples: AI-native email clients, AI-native document tools where creation is AI-assisted by default, AI-native CRM features where the CRM surface is becoming conversational.
The trend: most Category 1 tools (systems of record) are adding Category 3 interfaces. HubSpot, Salesforce, and QuickBooks are all adding AI-native features that let users query their data conversationally within the tool.
The question is whether those native AI features are good enough to replace the external AI layer; or whether the external AI layer, with the company’s context pack loaded, produces better results.
The conversational layer in practice: what it actually changes for a mid-market company today
The conversational layer thesis is not science fiction. It is already operational in the companies that have built the AI infrastructure described throughout this series.
The old way (dashboard and interface interaction):
On Monday morning, the ops lead opens the PM tool dashboard, filters for overdue tasks, exports to a spreadsheet, opens HubSpot to check pipeline status, navigates to the reporting view, notes the stalled deals, opens QuickBooks to check the cash position, navigates to the receivables report.
After 45 minutes: they have a picture of the business.
The conversational layer way:
The ops lead opens the AI workspace on Monday morning. The weekly briefing has already been generated; it synthesised the PM tool data, the pipeline data, and the cash position into a plain-language summary with flags.
They read it in three minutes and make three decisions.
Alternatively: the ops lead asks “which of our active projects have client-owned tasks that are more than seven days overdue?” The AI queries the PM tool data in real time and returns the list in 30 seconds. No filter building. No export.
What this requires:
- The AI to have access to the data in each tool (API connections built through Make or Zapier)
- The company context pack to be loaded (so the AI understands the company’s definitions, thresholds, and standards)
- The workflow or query to be either automated (the Monday briefing) or conversational (the ad-hoc question)
None of this is science fiction. It is buildable with current off-the-shelf tools. The companies doing it today built it in the previous six months.
The practical tool stack decisions: what to buy, what to consolidate, what to drop
Buy or prioritize:
- API-first systems of record. The CRM, accounting tool, and PM platform with the best API access are more valuable than ones with better interfaces. The interface is being replaced by the AI layer; the API is being used by it.
- The AI drafting and reasoning layer (Claude Teams, ChatGPT Team). This is the conversational layer itself. It is the most important tool investment in the AI-first stack.
- Workflow automation that connects the AI to the data sources (Make, Zapier, n8n). These are the connective tissue that makes the conversational layer operational.
Consolidate or replace:
- Standalone reporting and BI tools that primarily present data the systems of record already hold. When the AI layer can answer the questions these tools were built to answer, the standalone tool is redundant.
- Specializt “AI features” that duplicate what the AI layer already does better. Many SaaS tools are adding AI-powered summaries and insights that are less capable than a properly contextualised Claude or GPT-4 query. Do not pay for AI features in tools when the AI layer does the same thing better.
- Tools with no API. Any significant operational tool without API access is a liability in an AI-first stack; the data is trapped, invisible to the AI layer.
Do not drop yet:
- Systems of record, even if their interfaces are being replaced. The data still needs to live somewhere.
- Tools with strong API access and active development toward AI-native interfaces; these are becoming better, not worse.
Common questions on AI and the SaaS stack
”Should I stop buying new SaaS tools?”
No; but change the evaluation criteria. Before adding a new SaaS tool, ask: “Does this tool have an API that the AI layer can connect to?” A tool without API access is becoming a dead end. A tool with strong API access is an asset that compounds as the AI layer builds around it.
”Which SaaS tools are most at risk of being replaced by AI?”
The highest-risk category: tools whose entire value proposition is “a better view of your data from another tool.” Standalone reporting dashboards, analytics overlays, and read-only intelligence tools are being displaced by conversational AI queries that answer the same questions more flexibly. Tools that are systems of record (where the data is actually stored and updated) are not at risk.
”Does this mean I need fewer people to manage my tech stack?”
It means the nature of the work changes. The person who used to spend time navigating dashboards and compiling reports now spends that time making decisions. The tech stack management work; integration maintenance, API connections, workflow automation; becomes more important, not less. The skill set shifts from “knows how to use each tool” to “knows how to connect tools to the AI layer."
"How does this affect my company’s SaaS contracts if I want to switch?”
Most SaaS contracts are annual. The transition strategy: when contracts come up for renewal, evaluate whether the tool is a Category 1 (system of record, renew), Category 2 (interface/reporting layer, consider whether the AI layer has replaced it), or Category 3 (AI-native, accelerate adoption). Do not break contracts prematurely; evaluate at renewal.
”Which AI tools are becoming systems of record in their own right?”
The AI workspace (Claude Projects, ChatGPT custom GPTs) is becoming a system of record for the company’s AI context; the context pack, workflow documentation, and session history are data that matter. Treat the AI workspace with the same data governance as any other system of record; back it up, document it, and ensure the data is portable.
”How do I explain this shift to my CFO who just approved a multi-year SaaS contract?”
The multi-year SaaS contract for a Category 1 tool (CRM, accounting, PM) is likely fine; those tools are becoming more valuable, not less. The conversation to have with the CFO is about Category 2 tools; the reporting dashboards and analytics overlays that renewals might not justify if the AI layer is providing the same functionality more effectively.
Want the conversational layer built on top of the tools your company already runs on?
AI is not the new SaaS. It is the new interface above SaaS; a conversational layer that makes structured operational data queryable, synthesisable, and actionable through language rather than through point-and-click dashboards.
The SaaS tools that are systems of record become more valuable; the ones that are primarily interface or reporting layers become candidates for replacement.
For a mid-market company, the practical decisions are specific: prioritize API-first systems of record, invest in the AI orchestration layer, and begin consolidating the tools whose primary value is the interface that AI is now providing directly.
Path one: audit your current stack this week. List every SaaS tool the company pays for. Assign each to Category 1, 2, or 3. The Category 2 tools are the consolidation candidates; the renewals worth questioning.
Path two: bring in a partner. If you want the API connections, data integrations, and AI layer built above the existing tool stack so the conversational layer described in this article is operational; that is the work Phos AI Labs does in Phase 3. The team behind Phos AI Labs has helped 400+ businesses run on AI. The fastest way to know if it is the right fit is a conversation. Thirty minutes, no deck. Start here.