What companies are attacking expensive incumbent SaaS by rebuilding the workflow from scratch with AI?

The companies you’re describing are usually called AI-native vertical software, agentic applications, or SaaS challengers: startups that do not add a chatbot/copilot to an old workflow, but instead replace the workflow itself (often replacing multiple seats of an incumbent SaaS product with an AI operator). The most exposed incumbents tend to be expensive, seat-based systems where humans spend time moving information around, filling forms, reviewing documents, or coordinating tasks.

A few notable examples:

LeadPrysm SaaS Mag## Categories where this pattern is most active

1. Customer support → “AI employee”

Instead of selling 100 support seats, the new model is an agent that resolves tickets, updates systems, and escalates exceptions.

Examples:

• Sierra
• Decagon
• Intercom Fin AI

The incumbent question changes from “how do we make agents more productive?” to “how many human-agent workflows still need to exist?”

2. Legal → “AI associate”

Legal is attractive because much of the work is document-heavy and language-based.

Examples:

• Harvey
• Legora
• EvenUp

These companies are attacking workflows traditionally supported by expensive lawyers, paralegals, research platforms, and document tools.

3. Back office finance/accounting

This is one of the biggest targets because many processes are repetitive but expensive.

Targets:

• AP/AR automation
• bookkeeping
• procurement
• expense review
• financial close

Potential incumbents:

• Workday
• SAP
• Oracle
• Coupa Software

4. Recruiting and HR operations

The thesis: replace recruiter workflows, not just add AI search.

Targets:

• applicant tracking systems
• sourcing tools
• interview coordination
• screening workflows

5. Software engineering

The new product is not “a better IDE”; it is an AI engineer that owns tickets end-to-end.

Examples:

• Cognition
• Factory
• Cursor

The pattern investors are watching

The strongest candidates tend to have:

1. High labor cost

• A workflow currently requiring many people.

2. High software spend

• A company paying $50k–$5M/year for SaaS.

3. Structured inputs

• Documents, tickets, calls, forms, transactions.

4. Measurable outcomes

• Cases closed, claims processed, tickets resolved, revenue collected.

5. Weak system-of-record moat

• The incumbent stores data but does not own the actual business outcome.

The big strategic question is whether AI companies become a new layer on top of SaaS or whether they become the new system of action, with old SaaS becoming a backend database. Analysts and operators disagree on how broadly this will happen; some argue incumbents will successfully embed agents, while others expect more workflow replacement.

If you’re looking from a startup/founder angle, the most interesting “white spaces” are probably the ugly $100M–$10B SaaS categories that have lots of human operators, compliance burden, and document workflows (insurance, healthcare admin, construction, logistics, procurement, tax, and regulated finance).

The enterprise software landscape is undergoing a massive structural shift . Traditional Software-as-a-Service (SaaS)—built on the premise of giving humans a digital dashboard/tool to do work and charging them per seat—is being challenged by AI-native "systems of action" and agentic workflows . Instead of human users clicking through menus, logging entries, and generating reports inside an expensive legacy tool, AI-native platforms treat the LLM/agent as the primary user or worker, automating the end-to-end outcome rather than just assisting with a task.

Startups and scale-ups attacking expensive incumbents are focusing on several key operational domains:

• Customer Support & Service Operations: Instead of traditional ticketing systems like Zendesk or ServiceNow where human agents manually read, tag, route, and reply to tickets, AI-native alternatives (like Sierra , co-founded by Bret Taylor) are built from the ground up to autonomously converse, troubleshoot back-end database systems, and resolve entire customer workflows end-to-end.
• Sales, CRM & Revenue Operations: Traditional CRMs like Salesforce or HubSpot are databases meant for humans to manually input call notes, manage pipelines, and update deal stages. AI-first challengers (like Attio or emerging autonomous outbound and pipeline agents) rebuild the CRM to auto-capture telemetry, emails, and calendar metadata, treating the CRM as a living, autonomous system rather than a digital filing cabinet.
• Finance, Accounting & Billing: Incumbent ERP and billing tools (like NetSuite or legacy spend-management systems) require armies of accountants to reconcile invoices, process expenses, and run checks. AI-native financial platforms (such as Numeric for automated accounting close or various agentic AP/AR automation tools) ingest raw banking, invoice, and receipt data to execute the accounting workflow itself rather than just providing fields to type numbers into.
• Legal, Compliance & Contract Review: Traditional contract lifecycle management (CLM) software just stored PDFs and tracked signatures. AI-native legal platforms (like Disclo, Harvey , or specialized compliance agents) analyze, negotiate, redline, and risk-assess documents natively through legal-domain models, bypassing the slow, human-heavy workflows of legacy tools.
• Software Engineering & IT Ticketing: Traditional bug-tracking and project management tools (like Jira ) are being pressured by autonomous coding and QA agents (such as Devin by Cognition Labs or specialized engineering workflow platforms) that take a natural-language product requirement, write the code, run tests, and open the pull request directly, bypassing standard backlog-grooming workflows.

The New Playbook: Why Incumbents are Vulnerable

1. The Death of Per-Seat Pricing: Traditional SaaS thrives when a company buys 5,000 seats. AI-native competitors charge by the outcome, resolution, or token/compute usage , completely decoupling revenue from human headcount. If an AI agent does the work of 10 people, a per-seat model collapses the incumbent's revenue.
2. From System of Record to System of Action: Incumbents are mostly systems of record (where data sits passively). AI challengers build systems of action that actively ingest context, make API calls, trigger multi-step logic, and execute the business process without human data entry.
3. Data Layer Re-Architecture: True AI-native competitors don't just "bolt on a chatbot feature" to an old database schema; they rebuild the underlying data structures specifically to feed continuous context to autonomous agents.

Are you looking at this shift from an investment/market perspective , or are you trying to find specific AI-native alternatives to replace a bloated tool in your own tech stack? Let me know the category and I can narrow down the exact players.

Yes. The interesting cohort is not simply “AI SaaS startups.” It’s companies using AI to delete an incumbent’s workflow, often replacing a seat-based software product and some of the labor around it.

A useful framing is Bessemer’s “AI systems of action”: AI is weakening the traditional system-of-record moat around products like SAP, Salesforce, NetSuite and Workday because an agent can increasingly execute the workflow without a human operating the old UI.

Here are the companies I’d put on the shortlist:

The ones I think are most interesting

1. Sierra → customer support

This is probably the cleanest example of the thesis. Instead of selling another helpdesk UI, Sierra sells an agent that resolves the customer's issue. That's a fundamental change from “software that helps a support rep work” to “software that does the support work.”

Sierra reportedly reached $100M ARR in 2025 and was being used by companies including WeightWatchers, SiriusXM, Sonos and Redfin.

2. Harvey → legal

Harvey is attacking a particularly juicy combination: extremely expensive software + extremely expensive human labor. Rather than making lawyers marginally faster inside traditional legal software, it is automating research, drafting, analysis and other legal workflows.

That's why I consider Harvey more important than “AI legal research startup.” The ambition is closer to rebuilding the operating system for legal work.

3. Rogo → investment banking

Rogo is another strong example because banking has historically had enormous amounts of analyst labor wrapped around information systems. Its product handles things like presentations and financial models, and it is expanding from investment banking toward wealth management.

The interesting question isn't “Can Rogo beat Bloomberg?” It's how much of the workflow that caused a bank to buy Bloomberg, Excel, presentation software, research databases, etc. can disappear when the agent does the work?

4. Decagon → support

Decagon is slightly different from Sierra but follows the same thesis. The key metric isn't seats replaced; it's tickets/interactions resolved. One report says more than half of Decagon's customers replacing existing IVR, ticketing or CRM-based support systems moved to it.

That's the tell: the customer isn't merely adding AI to the stack. They're removing the old stack.

5. LayerX → back office

LayerX is particularly interesting because it's going after ugly, entrenched workflows—expenses, invoices, procurement, tax and HR. Its Bakuraku product competes with products including SAP Concur and other expense-management systems while using AI to automate the underlying processing.

This is exactly the kind of market where an AI-native company can ask:

Why does a human need to fill out this form at all? rather than:

How can we make the form slightly nicer?

---

The pattern I'd watch

The highest-potential targets have roughly this structure:

Expensive SaaS + lots of human labor + repetitive documents/data + terrible UX + per-seat pricing.

That's why legal, customer support, finance, recruiting, sales ops, compliance, insurance and healthcare administration are so interesting.

The startup doesn't necessarily need to replace the incumbent's database. It can initially replace the human interaction layer sitting on top of it.

That's the crucial distinction:

Old SaaS

Human → UI → workflow → database → outcome

AI-native

Human → intent → agent → outcome

The UI and much of the workflow become optional.

A 2026 BCG analysis similarly finds AI-native startups capturing disproportionate share in vertical and departmental software: startups account for 88% of the cited vertical-AI category and 60% of departmental AI spend.

And there's an even more aggressive version

The companies I would watch most closely are moving from:

“AI software replaces incumbent software”

to:

“AI service replaces the people who operate the software.”

That's the service-as-software model. Sierra is the canonical example: instead of selling customer-service software for $X per agent, the company can effectively sell resolved customer interactions. The software is becoming a delivery mechanism for labor.

That creates a much larger TAM because the startup isn't competing for the customer's $50k SaaS budget. It's competing for the $5M labor budget attached to the workflow.

So if you're looking for companies that could produce the next generation of $10B–$100B application businesses, I'd focus less on “AI features” and more on companies where the pitch sounds like:

“You don't need this software anymore. Tell us what you want done, and we'll do it.” That's the real attack on incumbent SaaS.

The software landscape is experiencing a massive shift from traditional "system of record" SaaS (where humans do the heavy lifting inside bloated, expensive dashboards) to AI-native agentic workflows (where autonomous digital workers execute the end-to-end task).

Rather than just adding a generic chatbot "copilot" to an existing product, these startups are completely bypassing the legacy user-interface and pricing model. Instead of charging per seat, they charge by outcome or task completion.

Prominent companies attacking expensive incumbent SaaS by rebuilding workflows from scratch include:

• 11x builds autonomous digital workers (like "Alice" for outbound sales development) that replace the traditional human SDR workflow. Instead of buying expensive sequence tools (like Outreach or SalesLoft) and paying multiple human seats to research, personalize, and email leads, 11x automates the entire funnel from raw market signals to booked meetings.
• Ambience Healthcare attacks legacy clinical documentation and electronic health record (EHR) bloat. Rather than forcing doctors to spend hours navigating expensive, rigid enterprise medical software, Ambience's AI operating system captures natural dialogue during patient visits to autonomously generate specialty-accurate notes, medical coding, and after-visit instructions.
• Artisan develops autonomous AI employees (dubbed "Ava") designed to act as digital sales and marketing team members. They fundamentally challenge traditional CRM-adjacent outreach and lead-generation tools by managing the workflow independently rather than just serving as a data-entry helper.
• AutoGenAI reimagines the arduous corporate bid-and-proposal writing workflow. Instead of enterprise teams spending hundreds of hours manually compiling documents and searching through old company repositories to respond to RFPs (Requests for Proposals), AutoGenAI builds the proposal natively using generative AI orchestration.
• Nabla provides ambient clinical intelligence that captures doctor-patient interactions to instantly populate clinical notes. It undercuts the administrative friction of legacy practice-management software by removing the data-entry bottleneck entirely.

Would you like to explore how these AI startups are shifting pricing models away from traditional per-seat SaaS, or look into how large legacy incumbents (like Salesforce and Workday) are trying to defend their moats?

The shift from traditional SaaS (where software is a tool humans log into to fill out forms, click buttons, and build dashboards) to AI-native workflow systems (where autonomous agents handle the execution and humans simply supervise) is redefining enterprise tech.

Instead of legacy incumbents trying to "bolt on" a chatbot feature, a new wave of AI-native disruptors is rewriting the underlying mechanics of customer support, legal ops, sales development, coding, and finance.

Here is how startups and emerging platforms are attacking expensive incumbent categories by rebuilding workflows from scratch:

• Customer Support & Helpdesks
  • The Incumbents: Zendesk, ServiceNow, Salesforce Service Cloud.
  • The AI-Native Attack: Traditional helpdesks are priced "per seat," charging companies more the more human agents they hire. AI-native alternatives like Sierra (founded by Bret Taylor and Clay Bavor) and autonomous support platforms (e.g., Decagon, Kapa.ai ) don't just route tickets to humans or suggest canned macros—they resolve end-to-end multi-step issues across backend databases, handle authentication, and process refunds autonomously. The economic model shifts from charging per human seat to paying per resolution, attacking the core unit economics of legacy helpdesks.
• Sales Prospecting & Outbound Development
  • The Incumbents: Outreach, Salesloft, ZoomInfo, HubSpot.
  • The AI-Native Attack: Traditional sales engagement platforms rely on humans building static sequences, manual list filtering, and template blasting. AI-native disruptors like 11x (building "Alice," an autonomous digital worker for sales development) and ** Artisan** (building "Ava") build the workflow around recursive, context-aware execution. They research prospects across the web in real-time, tailor hyper-personalized messaging based on recent company news or SEC filings, handle email replies natively, and book meetings without a human ever touching a sequence builder.
• Legal Contract Review & Compliance
  • The Incumbents: Ironclad, Thomson Reuters, traditional legal practice management software.
  • The AI-Native Attack: Legacy contract lifecycle management (CLM) software treats a contract as a static PDF stored in a digital filing cabinet with metadata tags applied by a paralegal. AI-native legal platforms like Harvey or Robin AI treat legal work as a continuous reasoning engine. They draft, negotiate, and redline contracts from scratch by simulating the company's precise risk playbook, drastically reducing reliance on outside counsel and legacy document-repository workflows.
• Software Engineering & Project Management
  • The Incumbents: Jira, Linear, GitHub (traditional issue tracking).
  • The AI-Native Attack: Traditional tools require humans to manually open tickets, log bugs, assign story points, and push status updates. Platforms leveraging agentic coding loops (e.g. , advanced implementations via Cursor, Devin by Cognition, and automated engineering managers) are shifting the workflow so that bugs auto-generate diagnostics, write their own pull requests, pass unit tests, and close the Jira/Linear ticket autonomously. The "project management" tool becomes secondary to the autonomous execution layer.
• Finance, Accounting & Expense Operations
  • The Incumbents: Ramp, Brex (which evolved rapidly), NetSuite, Concur.
  • The AI-Native Attack: While modern fintech incumbents have done a stellar job integrating software and cards, the pure accounting, reconciliation, and audit trail workflow is being re-engineered by autonomous AI bookkeeping agents. Startups in this space use multi-modal models to read messy, non-standardized invoices, categorize complex multi-subsidiary general ledger entries, and flag anomalies without human bookkeeping intervention.

Would you like to explore how the pricing models (per-seat vs. outcome-based) are shifting between these AI disruptors and legacy vendors, or do you want to look at a specific industry workflow (like HR or marketing)?

Yes. The interesting cohort isn’t “SaaS companies adding an AI assistant.” It’s AI-native companies trying to make the old SaaS category itself unnecessary—often by replacing a sequence of screens, humans, and workflows with an agent that simply gets the job done.

Redpoint calls the broader shift “SaaS → Service-as-Software”: instead of selling software that helps a person perform work, sell software that performs the work and charge for usage/outcomes. www.slideshare.net Gartner estimates that as much as $234B of enterprise-application spending could be exposed to this “agentic arbitrage” by 2030.

Here are the companies I'd put on the watchlist:

The ones I'd pay particular attention to

1. Sierra / Decagon — customer support

This may be the cleanest example of the thesis. Instead of selling a better Zendesk interface, they sell an AI agent that resolves the customer's problem. Redpoint specifically identifies Sierra and Decagon as leaders in this new support model.

The strategic difference is huge:

Old SaaS: “Give every support agent a seat.” AI-native: “Pay us for the customer issues we resolve.” That attacks both the product architecture and the pricing model.

2. Rillet / DualEntry / Campfire — ERP

This is probably the most interesting “incumbent SaaS replacement” battlefield because the incumbent products are extraordinarily expensive and painful to implement.

Rillet, DualEntry and Campfire are building AI-native accounting/ERP systems targeting NetSuite, Sage Intacct and QuickBooks. DualEntry's particularly interesting wedge is AI-assisted migration from legacy financial systems, attacking one of the biggest reasons customers don't switch.

Rillet is especially compelling because it isn't merely “AI bookkeeping”; it's trying to become the financial system of record.

3. Harvey — legal

Harvey is further along than most of these companies in turning AI from “software for lawyers” into software doing lawyer work. Redpoint highlights Harvey alongside Legora as examples of AI-native legal systems moving beyond traditional legal software toward actual work execution.

This is a particularly attractive market because the incumbent isn't just software—it's also hundreds of billions of dollars of human labor.

4. Clay — GTM

Clay is an interesting version where the target isn't one incumbent. It's attacking the stack of Salesforce + ZoomInfo + Clearbit + Apollo + Outreach + spreadsheets + human SDR research.

Its insight is essentially: why maintain six databases and make a salesperson orchestrate them when an agent can research the account, enrich it, generate personalization and execute the workflow?

Redpoint specifically calls out Clay and Unify GTM as examples of this new sales/marketing architecture.

5. Cursor — developer software

Cursor is slightly different because it attacks GitHub/Copilot-style tooling rather than classic business SaaS. But strategically it's perhaps the purest example of the phenomenon.

The old workflow was:

IDE → write code → search docs → Stack Overflow → run tests → debug → repeat

The new abstraction is increasingly:

“Build this.” → agent operates across the repository.

That's what makes AI-native applications dangerous to incumbents: they don't necessarily need to reproduce the incumbent's feature set. They can delete the workflow that made those features necessary.

The second-order opportunity is even more interesting

I think there are three tiers of this phenomenon:

1. AI feature — Salesforce adds an AI agent.
2. AI-native application — a startup rebuilds Salesforce around agents.
3. AI-native service — the customer doesn't buy CRM software at all; they buy the sales outcome.

The third is where things get really disruptive.

For example:

• Don't sell customer-support software → sell resolved tickets
• Don't sell accounting software → sell a closed, reconciled set of books
• Don't sell legal software → sell completed legal work
• Don't sell sales-engagement software → sell qualified pipeline
• Don't sell recruiting software → sell hires
• Don't sell insurance software → sell processed claims
• Don't sell AP software → sell invoices paid correctly

Redpoint's research explicitly describes this movement from SaaS toward “Service-as-Software”, with companies increasingly charging by usage or outcomes rather than seats.

And that's why I wouldn't define the opportunity as simply “AI-native SaaS.” The more interesting thesis is:

Take an expensive software category where the customer is really paying for a human workflow, then use AI to collapse the software + labor stack into an outcome. The strongest current examples, in my view, are Sierra/Decagon, Harvey, Rillet/DualEntry, Clay, Cursor, and Doss.

One caveat: this is still very much an emerging market. Incumbents have enormous integration/data advantages, and Gartner currently cautions that replacing core ERP wholesale with AI remains immature. Gartner So the most vulnerable categories are generally workflow-heavy applications, not deeply entrenched systems of record.

If you're thinking about this from a startup/investment angle, I can also map out 50+ incumbent SaaS categories and the specific AI-native startups attacking each one, ranked by how vulnerable the incumbent is.

This is probably the most interesting “incumbent SaaS replacement” battlefield because the incumbent products are extraordinarily expensive and painful to implement.

Rillet, DualEntry and Campfire are building AI-native accounting/ERP systems targeting NetSuite, Sage Intacct and QuickBooks. DualEntry's particularly interesting wedge is AI-assisted migration from legacy financial systems, attacking one of the biggest reasons customers don't switch.

Rillet is especially compelling because it isn't merely “AI bookkeeping”; it's trying to become the financial system of record.

3. Harvey — legal

Harvey is further along than most of these companies in turning AI from “software for lawyers” into software doing lawyer work. Redpoint highlights Harvey alongside Legora as examples of AI-native legal systems moving beyond traditional legal software toward actual work execution.

This is a particularly attractive market because the incumbent isn't just software—it's also hundreds of billions of dollars of human labor.

4. Clay — GTM

Clay is an interesting version where the target isn't one incumbent. It's attacking the stack of Salesforce + ZoomInfo + Clearbit + Apollo + Outreach + spreadsheets + human SDR research.

Its insight is essentially: why maintain six databases and make a salesperson orchestrate them when an agent can research the account, enrich it, generate personalization and execute the workflow?

Redpoint specifically calls out Clay and Unify GTM as examples of this new sales/marketing architecture.

5. Cursor — developer software

Cursor is slightly different because it attacks GitHub/Copilot-style tooling rather than classic business SaaS. But strategically it's perhaps the purest example of the phenomenon.

The old workflow was:

IDE → write code → search docs → Stack Overflow → run tests → debug → repeat

The new abstraction is increasingly:

“Build this.” → agent operates across the repository.

That's what makes AI-native applications dangerous to incumbents: they don't necessarily need to reproduce the incumbent's feature set. They can delete the workflow that made those features necessary.

The second-order opportunity is even more interesting

I think there are three tiers of this phenomenon:

1. AI feature — Salesforce adds an AI agent.
2. AI-native application — a startup rebuilds Salesforce around agents.
3. AI-native service — the customer doesn't buy CRM software at all; they buy the sales outcome.

The third is where things get really disruptive.

For example:

• Don't sell customer-support software → sell resolved tickets
• Don't sell accounting software → sell a closed, reconciled set of books
• Don't sell legal software → sell completed legal work
• Don't sell sales-engagement software → sell qualified pipeline
• Don't sell recruiting software → sell hires
• Don't sell insurance software → sell processed claims
• Don't sell AP software → sell invoices paid correctly

Redpoint's research explicitly describes this movement from SaaS toward “Service-as-Software”, with companies increasingly charging by usage or outcomes rather than seats.

And that's why I wouldn't define the opportunity as simply “AI-native SaaS.” The more interesting thesis is:

The thesis that AI-native startups are attacking expensive, seat-based legacy SaaS (Salesforce, Zendesk, ServiceNow) by completely rebuilding workflows around autonomous agents rather than human-driven data entry is reshaping the software landscape. Instead of charging per user seat, these companies often price based on outcomes, resolution rates, or consumed compute.

Prominent AI-first challengers across key enterprise categories are redesigning traditional SaaS workflows from the ground up:

Customer Support & Success (Replacing Zendesk, Intercom, ServiceNow)

Legacy helpdesks were built as ticketing systems for humans to read, triage, and manually reply to queries. AI-native alternatives treat the support queue as an autonomous resolution problem.

• Sierra: Co-founded by Bret Taylor and Clay Bavor, Sierra builds conversational AI agents that handle end-to-end customer interactions with deep business logic, context integration, and brand voice, pressing hard on traditional enterprise customer service footprints.
• Decagon: Focuses on autonomous, highly complex customer support agents tailored for large enterprises, competing directly with incumbent service clouds by completely bypassing the traditional "agent inbox" workflow.
• Kapa.ai: Rebuilds technical developer support and documentation workflows into real-time, context-aware expert assistants rather than static knowledge bases and ticketing queues.

Sales & Revenue Operations (Replacing Salesforce, Outreach, HubSpot)

Traditional CRM workflows require humans to log calls, update pipelines, and manually trigger sequences. AI-native tools treat the CRM as a continuous background data layer driven by ambient listening and autonomous outreach.

• 11x: Builds digital workers (AI "employees" like Alice for sales development) that execute end-to-end outbound prospecting, research, and engagement workflows without a human needing to click through CRM lists or sales engagement sequences.
• Bland AI: Rebuilds phone-based enterprise communication workflows (sales, collections, qualification) via hyper-realistic conversational voice agents, replacing entire call-center software stacks.

Legal, Compliance, and Finance Back-Office (Replacing Legacy ERP/Legal Tools)

Document-heavy, highly manual processes that previously required armies of analysts or expensive point-solution SaaS are being collapsed into single-prompt or autonomous continuous workflows.

• Harvey: Rebuilds legal workflows (contract analysis, litigation prep, due diligence) from the ground up using specialized legal LLMs, challenging legacy legal tech and massive billable hours.
• Numeric / Vic.ai: Rebuilds accounting and financial-close workflows by utilizing AI to autonomously handle bookkeeping, invoice processing, and reconciliations, bypassing traditional heavy ERP data-entry modules.

Would you like to explore how the pricing models (seat-based vs. outcome-based) are shifting for these AI-native challengers, or look into how traditional incumbents like Salesforce and ServiceNow are fighting back?

Yes. The interesting cohort is not “SaaS companies with AI features.” It’s companies asking: if an AI agent is the primary user, why reproduce the 1990s/2000s workflow at all?

That creates a fairly specific attack vector against expensive, sticky incumbents: take the budget of an incumbent system of record, eliminate much of the human-operated workflow, and charge for outcomes/usage instead of seats. Recent industry analysis makes the same point: document-heavy + workflow-heavy + seat-priced software is especially exposed.

Here are the companies I'd put on the watchlist.

The strongest examples

But there are some important distinctions.

1. Serval → ServiceNow

Probably the cleanest example of the thesis.

Serval is explicitly going after ServiceNow rather than merely becoming another AI add-on.

Its Catalyst agent analyzes ticket history, determines what can be automated, and generates the underlying code/workflow/permissions for approval. The architecture is built around code-generation agents rather than a traditional workflow-builder bolted onto an old ITSM system. Serval has raised $127M and reached a reported $1B valuation.

The really interesting economic attack is:

ServiceNow: humans operate the workflow inside the system. Serval: the system operates the workflow. That's a much bigger disruption than “AI-powered ServiceNow.”

2. Rillet / DualEntry / Campfire → NetSuite

This may be the most developed “AI-native ERP” battleground.

Rillet is rebuilding the accounting system around continuous processing rather than the traditional month-end-close workflow. Its pitch is essentially: don't make accountants operate an ERP; have the ERP continuously perform the accounting.

DualEntry attacks another enormous incumbent pain point: migration itself. Its “NextDay Migration” uses AI to map and migrate historical financials from legacy systems, directly attacking one of the biggest reasons companies remain trapped in NetSuite.

Campfire is going even broader: finance/ERP as a new operating system rather than a collection of modules.

The category is unusually interesting because the incumbents have enormous implementation costs and administrative overhead. The challengers can potentially turn:

ERP → database humans maintain

into:

ERP → autonomous financial system.

3. qomplement → SAP / NetSuite

qomplement is an especially aggressive example.

Its thesis is to start with a painful operational workflow—procurement, inventory, freight, ERP updates—and progressively replace the underlying ERP. YC describes it explicitly as an agentic ERP designed for AI agents rather than human users, with a goal of replacing SAP and NetSuite.

This is the “don't replace the whole ERP on day one” strategy:

1. automate one ugly workflow,
2. become the system of action,
3. absorb adjacent workflows,
4. eventually become the ERP.

That's potentially much more viable than asking a CFO to rip out SAP on day one.

4. Sierra → Salesforce / Zendesk / contact-center software

Sierra is another extremely important one.

Rather than selling a better agent-assist tool, Sierra sells customer-service agents that actually execute the business process: refunds, insurance claims, refinancing, order returns, etc.

As of May 2026, Sierra said it was serving more than 40% of the Fortune 50 and that its agents were powering billions of customer interactions. It had raised $950M at a valuation above $15B.

The conceptual shift is:

Zendesk: manage tickets.

Sierra: resolve the customer's problem.

That distinction matters enormously for pricing. If AI eliminates the human ticket handler, per-agent/per-seat pricing becomes economically nonsensical.

5. Harvey / Legora / Manifest → legal SaaS

Legal is probably the most obvious professional-services category where the workflow itself can be reconstructed.

Harvey has gone well beyond “legal research chatbot.” It is becoming an operating layer for legal work—research, drafting, review, analysis, etc. Recent reporting puts Harvey at roughly $350M ARR and a $11B valuation.

But Manifest is arguably the purer “rebuild the business” thesis.

Its AI-native law-firm model combines intake, drafting, billing, review and quality control into an operating model based on fixed fees rather than traditional billable hours.

That's fascinating because it's attacking both software and the business model underneath the software.

Instead of:

law firm + associates + legal SaaS + hourly billing the new model can become:

AI workflow + supervising lawyers + fixed-price outcome.

6. Decagon → Zendesk / Salesforce Service Cloud

Decagon is another serious one.

The important distinction versus classic “AI help desk” startups is that the product is increasingly about autonomous resolution, not helping a support agent type faster.

That's the pattern I'd watch: once an agent can resolve 70–90% of interactions, the incumbent ticketing UI becomes increasingly like a legacy database sitting underneath the actual customer experience.

Sierra is going after the enterprise version of this; Decagon is another major manifestation of the same thesis. Current market comparisons explicitly place Decagon among the AI-first challengers to Zendesk.

7. AI-native CRM is coming, but it's less mature

This is a fascinating category because CRM is arguably one of the most vulnerable pieces of SaaS architecture.

The traditional CRM workflow is:

email → human enters data → human updates fields → human moves opportunity → manager reads dashboard.

AI can instead observe communications and continuously construct the commercial graph itself.

Companies worth watching include:

• Attio — modern CRM/data model with AI-native attributes.
• Day AI — attempts to create the CRM record from communications rather than requiring humans to maintain it.
• Reevo — pushes toward consolidating GTM workflow.
• Monaco — goes even further toward selling the sales operation rather than CRM software.

Attio is already being positioned as an AI-native alternative to Salesforce, although I'd classify it as modern CRM with an AI-native architecture, rather than as radical a replacement thesis as Serval or Rillet.

The pattern I'd actually bet on

If you're trying to identify the next generation of companies that can kill $100M–$1B SaaS categories, I wouldn't primarily look for “AI SaaS.”

I'd screen for this combination:

1. Expensive incumbent $20K–$1M+ annual contracts.

2. Human-mediated workflow Lots of people clicking, copying, reviewing, reconciling or routing things.

3. High implementation cost The incumbent's complexity is partly its moat.

4. AI can perform the actual work Not merely summarize it.

5. Agent can become system of action The new product actually does things in other systems.

6. Data exhaust becomes the moat Every completed workflow improves the agent.

7. Pricing can move from seats → outcomes Per resolution, transaction, claim, close, case, shipment, etc.

That produces a much more interesting map:

• ServiceNow → Serval
• NetSuite → Rillet / DualEntry / Campfire / qomplement
• Salesforce Service Cloud / Zendesk → Sierra / Decagon
• Thomson Reuters / LexisNexis → Harvey
• Traditional legal-services stack → Manifest
• FloQast / BlackLine → Numeric
• Salesforce CRM → Day AI / Reevo / Attio
• Vertical ERP → a rapidly growing cohort of agentic vertical systems

And the broader market is starting to validate this architecture shift: BCG's 2026 analysis found AI-native startups substantially further along the AI product-maturity curve than incumbent SaaS, with startups already capturing the majority of startup spend in vertical AI and a majority in departmental AI.

My shortlist of the 10 I'd watch most closely

If the question is “who might actually take meaningful budget away from incumbent SaaS?”, rather than simply “who has a cool AI product?”, my list would be:

1. Serval — ServiceNow
2. Rillet — NetSuite
3. Sierra — Salesforce/Zendesk/contact centers
4. DualEntry — NetSuite/accounting
5. Harvey — legal software
6. Campfire — ERP/finance
7. qomplement — SAP/NetSuite/supply chain
8. Decagon — customer support
9. Manifest — law-firm operating model
10. Numeric — financial close

The really important signal isn't that these companies use LLMs. It's that they're asking a more dangerous question of the incumbents:

“Why does this workflow need to exist in its current form at all?” That's the distinction between an AI feature and an AI-native SaaS replacement.

5. Agent can become system of action The new product actually does things in other systems.

6. Data exhaust becomes the moat Every completed workflow improves the agent.

But Manifest is arguably the purer “rebuild the business” thesis.

Its AI-native law-firm model combines intake, drafting, billing, review and quality control into an operating model based on fixed fees rather than traditional billable hours.

That's fascinating because it's attacking both software and the business model underneath the software.

Instead of:

law firm + associates + legal SaaS + hourly billing the new model can become:

AI workflow + supervising lawyers + fixed-price outcome.

6. Decagon → Zendesk / Salesforce Service Cloud

Decagon is another serious one.

The important distinction versus classic “AI help desk” startups is that the product is increasingly about autonomous resolution, not helping a support agent type faster.

That's the pattern I'd watch: once an agent can resolve 70–90% of interactions, the incumbent ticketing UI becomes increasingly like a legacy database sitting underneath the actual customer experience.

Sierra is going after the enterprise version of this; Decagon is another major manifestation of the same thesis. Current market comparisons explicitly place Decagon among the AI-first challengers to Zendesk.

7. AI-native CRM is coming, but it's less mature

This is a fascinating category because CRM is arguably one of the most vulnerable pieces of SaaS architecture.

The traditional CRM workflow is:

email → human enters data → human updates fields → human moves opportunity → manager reads dashboard.

AI can instead observe communications and continuously construct the commercial graph itself.

Companies worth watching include:

The software industry is undergoing a structural shift from selling seats and dashboards (traditional SaaS) to selling completed work and autonomous execution (AI agents/services-as-software). Instead of forcing humans to click through multi-step forms, tabs, and export CSVs, AI-native challengers are rethinking workflows from the ground up around intent and automated outcomes.

The companies and categories aggressively attacking expensive incumbents fall into a few clear trenches:

1. Customer Service & Support Operations

• The Incumbents: Zendesk, Salesforce Service Cloud, ServiceNow.
• The AI-Native Challengers: Decagon, Sierra , and Gradient Labs.
• How they rewrite the workflow: Traditional helpdesks charge per human agent "seat" and treat AI as a mere chatbot deflection tool on top of a ticketing queue. AI-native platforms like Decagon treat the AI as the primary worker, capable of handling complex end-to-end resolution paths (issuing refunds, navigating internal logistics APIs, executing backend database actions) rather than just logging tickets for humans to fix.

2. Autonomous Software Engineering & QA

• The Incumbents: Jira, GitHub (traditional project management), legacy QA testing suites.
• The AI-Native Challengers: Devin (Cognition Labs), Cursor , and automated QA/testing agents.
• How they rewrite the workflow: Instead of humans writing tickets in Jira, assigning them to developers, reviewing PRs, and manually tracking velocity, these tools take high-level product specifications or bug reports and autonomously plan, write, test, and debug code end-to-end. The "project management dashboard" becomes secondary to the agent executing the engineering labor itself.

3. Legal, Compliance, & Contract Processing

• The Incumbents: Legacy contract lifecycle management (CLM) like Ironclad or DocuSign, expensive outside counsel workflows.
• The AI-Native Challengers: Legal AI platforms and specialized reasoning agents that draft, redline, and negotiate NDAs or vendor contracts based on custom corporate playbooks from second zero.
• How they rewrite the workflow: Traditional CLMs are filing cabinets and routing tools. AI-native legal platforms act as the first-chair reviewer, executing the cognitive analysis and negotiation markup directly rather than just storing the PDF.

4. Finance, Accounting, & Accounts Payable

• The Incumbents: NetSuite, Workday, Coupa.
• The AI-Native Challengers: Autonomous accounting and invoice-processing startups that ingest unstructured financial data, reconcile ledgers, handle cross-border compliance, and flag anomalies without requiring a massive team of human analysts to data-entry match receipts.
• How they rewrite the workflow: Moving away from ERP interfaces designed for human data entry to continuous, agentic back-office execution where software performs the audit and reconciliation natively.

The Two Strategic Plays: "Replace" vs. "Overlay"

When looking at how these companies attack incumbents, notice two distinct tactical approaches:

1. The Overlay / Trojan Horse Strategy: Companies like Intercom Fin or eesel AI plug directly into existing legacy systems (Zendesk, Salesforce, HubSpot). They capture the workflow value first by performing the actual work on top of the incumbent's database, starving the incumbent of their seat-based utility.
2. The Rip-and-Replace Strategy: Deeply vertical AI agents that don't care about the legacy system of record at all because they generate the outcome directly for the business buyer (charging per resolution, per output, or per completed task rather than per user license).

If you want to zoom in, let me know:

• Are you looking at a specific vertical (e.g., healthcare, fintech, legal)?
• Are you interested in the venture capital / market dynamics or specific technical architectures (like multi-agent systems)?

The companies I’d put in this category are AI-native workflow rebuilders: they are not selling “AI features” inside an existing SaaS product; they are trying to replace the process that the SaaS product encoded. The common pattern is:

Old SaaS: “A human uses software to complete a workflow.” AI-native: “An agent completes the workflow and humans supervise exceptions.” This is especially threatening where incumbents are expensive, seat-based, and built around repetitive knowledge work.

Legal (attacking Thomson Reuters / RELX / legal workflow SaaS)

• Harvey — Probably the clearest example. It is rebuilding legal work around AI agents: contract review, research, memo drafting, due diligence, and litigation workflows rather than being another legal database.
• Legora — Focuses on AI-native legal workflows for law firms and corporate legal teams.
• Robin AI — Contract review and contract lifecycle workflows rebuilt around AI.

Thesis: The winner may not be the company with the best legal search engine; it may be the company that owns “get this contract negotiated and signed.”

---

Customer support (attacking Zendesk / Salesforce Service Cloud)

• Decagon — Builds AI agents that resolve customer issues end-to-end.
• Sierra — AI agents designed to handle customer conversations and transactions for enterprises.
• Intercom — More of an incumbent AI transition case, but its AI agent push shows the category shift.

Thesis: A ticketing system is less valuable if an AI agent closes the ticket without a human touching it.

---

Sales / CRM (attacking Salesforce)

• Clay — Reimagines prospecting as AI-driven research, enrichment, and outbound generation.
• 11x — Builds autonomous AI sales workers rather than sales productivity software.
• Artisan — AI SDR-style agents replacing parts of outbound sales workflows.

Thesis: CRM historically records what salespeople do. AI-native companies aim to do the selling.

---

Recruiting / HR (attacking Workday / Greenhouse Software)

• Mercor — Uses AI interviews and matching to automate recruiting processes.
• Eightfold AI — Uses AI for talent matching and workforce workflows.
• Paradox — Automates candidate interactions and recruiting workflows.

Thesis: ATS systems manage candidates; AI recruiters may find, screen, and schedule them.

---

Finance / accounting (attacking Intuit / ERP workflows)

• Ramp — Moving from expense software toward autonomous finance operations.
• Brex — Similar move toward AI-managed finance workflows.
• Numeric — Automates accounting close workflows.

Thesis: The future accounting system may not be a ledger UI; it may be an agent that reconciles, explains, and closes books.

---

Engineering / developer tools (attacking Atlassian / traditional dev tooling)

• Anysphere — Creator of Cursor, an AI-native coding environment. McKinsey has highlighted Cursor and similar companies as examples of AI-native products built around new workflows rather than legacy software patterns.
• Replit — Makes software creation accessible through AI agents.
• Cognition — Builds AI software engineers such as Devin.

Thesis: Jira tracks engineering work; AI agents may perform much of the work.

---

Back-office / vertical operations

• Norm AI — Targets compliance workflows.
• Taktile — Automates regulated decision workflows.
• Async — Rebuilds SMB operations around AI agents rather than adding another SaaS dashboard.

---

The startups I’d watch most closely

If the question is “who has the strongest SaaS replacement thesis?” rather than “who has the best AI feature?”, I’d watch:

1. Harvey — legal workflows
2. Sierra — customer operations
3. Decagon — support automation
4. Cognition / Cursor — software engineering
5. Clay — sales automation
6. Ramp — finance operations
7. Mercor — recruiting
8. Norm AI / Taktile — regulated workflows

The biggest targets are categories where:

• the incumbent charges per seat,
• humans spend hours moving information between systems,
• the workflow is repetitive but requires judgment,
• the “database” matters less than the “action.”

A useful mental model: Salesforce, Workday, SAP, ServiceNow, etc. became systems of record. The AI-native challengers are trying to become systems of execution.