Yes. The interesting category is not “AI features for SaaS.” It’s companies asking: if we were building this workflow today, with agents able to do the work, would we build the same application at all?

That distinction matters. Gartner estimates as much as $234B of enterprise application software spend is exposed to “agentic arbitrage” by 2030, because agents can execute work across systems without users living inside traditional SaaS interfaces.

Here’s my current shortlist of companies I’d watch.

The strongest examples

1. Rillet / DualEntry / Campfire are probably the clearest thesis

AI-native ERP is one of the cleanest examples because legacy ERP has enormous amounts of expensive human coordination wrapped around it.

The current crop explicitly positions itself against NetSuite and Sage Intacct:

• Rillet — AI-native GL/ERP for mid-market companies.
• DualEntry — particularly interesting because AI is being used to attack the migration/implementation moat, historically one of ERP's biggest barriers to displacement.
• Campfire — broader finance suite spanning GL, revenue, close, billing and treasury.
• Light — European/global multi-entity finance.
• Doss — attacking the operational side of ERP.

An independent 2026 survey of the category describes “switch off NetSuite” as essentially the sales pitch and identifies Rillet, DualEntry, Campfire and Light as the leading AI-native ERP cohort.

Rillet is especially notable: it raised $100M at a $1B valuation in 2026 and has 600+ customers, while explicitly challenging Oracle/NetSuite-style ERP.

---

2. Von is the “burn the ships” example

This one is conceptually fascinating.

Rattle originally built conventional sales software. Then the founders realized that putting AI onto the existing product was fundamentally wrong. They cut the company from ~70 people to ~15, stopped selling the original product, and rebuilt around an autonomous revenue agent.

The resulting company, Von, operates more like an AI teammate sitting across the sales stack:

• updates CRM records
• identifies opportunities
• monitors deals
• forecasts revenue
• executes sales/revops tasks

It recently reached about $1M ARR.

That's almost the purest expression of your question:

Don't build a better Salesforce UI. Build the thing that makes Salesforce unnecessary.

---

3. Legal is another huge attack surface

Harvey and EvenUp are compelling because legal software has enormous amounts of human labor embedded between the database and the final outcome.

The old stack looks like:

database → search → documents → lawyer → analysis → drafting → review → client

The AI-native version is closer to:

case → agent → finished legal work

That's potentially much more economically powerful than “AI-powered legal search.”

EvenUp is particularly interesting in personal injury because it isn't trying to be a general-purpose legal copilot; it is rebuilding a specific, highly structured workflow around AI.

---

4. Customer support may be the most obvious SaaS category to get eaten

The traditional model:

customer → ticket → Zendesk → human agent → knowledge base → macros → escalation

The AI-native model:

customer → agent → resolution

That's why I'd put Sierra and Decagon high on the list.

The important strategic difference is that the AI company isn't necessarily trying to sell a better ticketing system. It's trying to make the ticketing system irrelevant.

Interestingly, even incumbents are acknowledging the direction: Zendesk is moving toward AI agents that resolve employee-service requests conversationally rather than simply improving its traditional interface.

---

5. Clay is the GTM version of the same idea

Clay is slightly different because it isn't simply “AI ZoomInfo.”

Its deeper thesis is:

Why should a salesperson operate five GTM databases and manually stitch together research, enrichment and outreach? Instead, AI can dynamically research a company/person, assemble context, enrich data, determine the appropriate action and trigger the workflow.

That's a much more fundamental rethinking of the sales stack. Clay's growth has also been tied to the emergence of the “GTM engineer” role—the person who operates these AI-driven systems rather than traditional sales tooling.

---

The pattern I'd use to find the next ones

I'd look for markets with these characteristics:

1. $20k–$500k/year incumbent software
2. Lots of humans operating the software
3. Workflow consists largely of unstructured inputs
4. Lots of copying/pasting between systems
5. Expensive implementation and customization
6. Outcome is more valuable than the UI
7. Incumbent pricing is based on seats
8. An AI agent can plausibly own the workflow

That produces some particularly interesting hunting grounds:

And there's a subtle but important point: the biggest opportunities may not look like SaaS replacements initially.

Bain estimates roughly $100B of U.S. SaaS opportunity could emerge from agents automating the coordination work between existing systems—people pulling data from ERP, reconciling spreadsheets, interpreting emails, making decisions, and escalating exceptions.

So there are really two plays:

A. “Replace Salesforce.” Build a better AI-native system of record.

B. “Make Salesforce irrelevant.” Build an agent that performs the job Salesforce currently helps a human perform.

I think B is the more radical—and potentially much larger—thesis.

If you're looking at this from a venture-investing / startup-idea perspective, my highest-conviction buckets right now would be ERP/accounting, insurance claims, legal, procurement, healthcare administration, and property management. The common denominator is not “lots of software.” It's lots of expensive humans moving information through expensive software to produce a relatively standardized business outcome.

---

4. Customer support may be the most obvious SaaS category to get eaten

The traditional model:

customer → ticket → Zendesk → human agent → knowledge base → macros → escalation

The AI-native model:

customer → agent → resolution

That's why I'd put Sierra and Decagon high on the list.

The important strategic difference is that the AI company isn't necessarily trying to sell a better ticketing system. It's trying to make the ticketing system irrelevant.

Interestingly, even incumbents are acknowledging the direction: Zendesk is moving toward AI agents that resolve employee-service requests conversationally rather than simply improving its traditional interface.

---

5. Clay is the GTM version of the same idea

Clay is slightly different because it isn't simply “AI ZoomInfo.”

Its deeper thesis is:

Why should a salesperson operate five GTM databases and manually stitch together research, enrichment and outreach? Instead, AI can dynamically research a company/person, assemble context, enrich data, determine the appropriate action and trigger the workflow.

That's a much more fundamental rethinking of the sales stack. Clay's growth has also been tied to the emergence of the “GTM engineer” role—the person who operates these AI-driven systems rather than traditional sales tooling.

---

The pattern I'd use to find the next ones

I'd look for markets with these characteristics:

1. $20k–$500k/year incumbent software
2. Lots of humans operating the software
3. Workflow consists largely of unstructured inputs
4. Lots of copying/pasting between systems
5. Expensive implementation and customization
6. Outcome is more valuable than the UI
7. Incumbent pricing is based on seats
8. An AI agent can plausibly own the workflow

That produces some particularly interesting hunting grounds:

The software industry is undergoing a structural shift from traditional "Software-as-a-Service" (characterized by human-driven dashboards, clicks, and seat-based pricing) to "Services-as-Software" or AI-native agentic workflows . Instead of building a tool for a human to do a job, these new companies build an AI agent that does the job, completely bypassing legacy workflows.

This disruption targets expensive incumbents across customer support, sales, engineering, finance, and legal operations.

1. Customer Support & CX

Legacy giants like Zendesk and Intercom rely on ticketing systems, macros, and human agents. AI-native alternatives are rebuilding support by letting autonomous agents resolve 80%–90% of end-user issues natively.

• 11x.ai: Builds digital workers (like "Alice") that act as fully autonomous sales development reps (SDRs) or support agents, replacing the traditional seat-based CRM/engagement workflow.
• Bland AI: Replaces traditional call-center software and complex IVR systems by building conversational, ultra-realistic voice agents that handle end-to-end phone workflows without human intervention.
• Kapa.ai / Eesel AI: Replaces rigid documentation and static help-desk tools by integrating deeply with technical knowledge bases to provide autonomous, context-aware conversational resolution.

2. Sales & Revenue Operations (RevOps)

Incumbents like Salesforce and HubSpot require manual data entry, pipeline updating, and strict CRM field management.

• Attio / Clay: While blending data enrichment and CRM, tools like Clay have re-architected how go-to-market teams build prospect lists—moving away from static database subscriptions (ZoomInfo) to dynamic, AI-scripted data waterfalls.
• Regie.ai / Artisan: Rebuilds outbound sales workflows by substituting manual sequences and templates with autonomous AI agents that research, personalize, and execute multichannel outreach.

3. Engineering & IT Operations

Legacy incident-management and ticketing tools like Jira or ServiceNow are being pressured by platforms that diagnose and fix issues automatically.

• Devin (Cognition Labs) & Plandek/Cursor ecosystems: Shifting software development workflows from human-driven project tracking to AI engineering teammates that take a prompt and execute multi-file code changes, testing, and deployment autonomously.
• Mend, Datadog-adjacent AI agents: Automating root-cause analysis rather than forcing engineers to sift through complex dashboard metrics manually.

4. Finance, Legal & Back-Office Workflows

Incumbents like Workday, NetSuite , or legacy document-review tools are facing attacks from unstructured-to-structured AI processing engines.

• Numeric / Vic.ai: Rebuilding accounting and close processes by using AI to autonomously ingest, categorize, and reconcile financial data, reducing reliance on massive manual ERP bookkeeping tiers.
• Harvey AI: Rebuilds legal workflows (contract analysis, drafting, and regulatory research) from the ground up using specialized LLMs rather than forcing lawyers through traditional document-management portals.

Key Themes of the Shift

• Outcome-Based vs. Seat-Based Pricing: Traditional SaaS charges per human "seat." AI-native builders are moving toward consumption or outcome-based pricing (e.g., paying per resolved ticket or successful outreach), which directly undermines the multi-seat pricing models of legacy giants.
• From System of Record to System of Action: Legacy SaaS merely stores data that humans input. AI-native tools close the loop by performing the final action natively via APIs and multi-step reasoning.

If you want to zoom in on a specific industry, let me know which sector you're tracking (Sales/CRM, Customer Support, Finance/Legal , or Engineering ), and I can map out more niche insurgents and how incumbents are scrambling to respond.

Yes. The pattern you’re describing is increasingly being called AI-native / “SaaS challenger” software: instead of adding a copilot to a legacy product, the startup rebuilds the underlying workflow around agents, often collapsing dozens of human steps into an outcome. YC and Bessemer have explicitly highlighted this as a major opportunity.

Here are the companies I’d watch most closely as of August 2026:

The clearest “replace the incumbent” bets

1. Rillet → ERP

This is probably the cleanest example of the thesis.

Rillet isn't merely putting an AI assistant on top of an accounting system. Its ledger, accounting workflows and agents are designed together. The company says more than 600 customers now use it, with customers replacing NetSuite, Oracle, SAP, Workday and other legacy systems.

The interesting insight is that the old ERP workflow is essentially:

human enters data → human reconciles → human reviews → human closes books. The AI-native version becomes:

systems generate data → agents reconcile/post/explain → humans approve exceptions. That's a fundamentally different product.

And Rillet isn't alone. DualEntry, Campfire and Light are part of an emerging AI-native ERP cluster explicitly going after NetSuite/Sage Intacct/QuickBooks.

2. Harvey + Legora → legal software

Legal is probably the other spectacular example.

Harvey is moving from “AI legal assistant” toward an operating layer where agents execute research, drafting, review, diligence, compliance and other legal workflows. Harvey reached an $11B valuation in March 2026.

Legora is the other company I'd put on the short list. It's attacking the same basic premise from a somewhat different product direction and has rapidly accumulated law-firm customers.

The important distinction is that these aren't necessarily trying to become “better LexisNexis.” They're trying to make the lawyer's old sequence of research → retrieve → compare → draft → revise → cite → review disappear.

3. Sierra + Decagon → customer support

This category is particularly interesting because the incumbent's basic unit is a ticket and a human seat.

Sierra is pushing toward the opposite abstraction: the resolved customer problem. Sierra explicitly describes its newer agents as orchestrating interactions over days or weeks and says its model is outcomes-based rather than token-based.

Decagon is another serious entrant. The 2026 competitive landscape increasingly divides the market into Zendesk/Intercom/Salesforce incumbents versus agent-native companies such as Sierra and Decagon.

That's a powerful wedge because an AI-native company can say:

“Don't buy 500 support seats. Pay us to resolve 2 million customer interactions.”

The incumbent has a structural problem: the better its AI gets, the fewer seats it needs to sell.

4. The next wave: vertical software

This is where I think the thesis gets especially interesting.

The strongest targets aren't necessarily giant horizontal products. They're expensive vertical SaaS where the software is basically a database wrapped around a labor-intensive workflow.

Think:

• insurance claims
• mortgage underwriting
• healthcare administration
• compliance
• tax
• payroll
• procurement
• construction management
• property management
• freight/logistics
• clinical administration
• recruiting
• government contracting

The winning startup doesn't necessarily say:

“Here's an AI feature for your existing software.” It says:

“Why are you buying software for this workflow at all? We'll just do the workflow.” That's the jump from SaaS → agentic software → AI-native service.

A particularly interesting boundary: Clay

Clay is a useful example because it sits between the old and new models.

Traditional GTM software gives a salesperson databases, CRM fields, sequences, filters and dashboards.

Clay increasingly gives the user an objective and lets AI assemble the research, enrichment, scoring and outbound workflow.

So instead of:

Salesforce + ZoomInfo + Apollo + enrichment tools + sequencing + spreadsheets + human research the AI-native product can increasingly become:

“Find me the 500 companies most likely to buy this, figure out who I should talk to, research them, and generate the outreach.” That's the same underlying phenomenon, even if Clay isn't literally replacing Salesforce today.

---

The companies I'd put in the top tier

If I were building an investment watchlist specifically around “expensive incumbent SaaS gets rebuilt from scratch by AI”, I'd rank the thesis roughly:

1. Rillet — strongest demonstrated ERP replacement signal
2. Harvey — strongest legal workflow attack
3. Sierra — strongest customer-service “outcome vs seat” thesis
4. Legora — formidable Harvey competitor
5. DualEntry — interesting ERP migration wedge
6. Campfire — another serious NetSuite challenger
7. Decagon — agent-native support
8. EvenUp — vertical legal workflow → outcome
9. Clay — GTM workflow reconstruction
10. Glean — enterprise knowledge/work orchestration

The common denominator isn't really AI.

It's removing the human-shaped architecture of SaaS.

Legacy SaaS asks:

“Which screens does the employee need to operate?” AI-native software asks:

“What outcome does the company want, and how much of the work can the system perform itself?” That shift also explains why seat-based pricing is vulnerable. An AI-native entrant can charge per transaction, outcome, resolution, filing, close, claim or completed workflow. Incumbents often can't do that without cannibalizing their own seat revenue.

And there is now evidence this isn't just an investor narrative: companies are actually ripping out incumbent systems. Rillet says roughly half of its customers migrated from legacy ERP/accounting products, with customers specifically replacing Intuit, NetSuite, Sage Intacct and Oracle.

If you're looking at this from a venture/investment perspective, I'd actually narrow the universe further: there are ~10–20 categories where the combination of high ACV + ugly implementation + repetitive human workflow + seat pricing + weak network effects makes an incumbent unusually vulnerable. That's where the really interesting “AI eats SaaS” companies are likely to emerge.

the AI-native product can increasingly become:

“Find me the 500 companies most likely to buy this, figure out who I should talk to, research them, and generate the outreach.” That's the same underlying phenomenon, even if Clay isn't literally replacing Salesforce today.

---

The companies I'd put in the top tier

If I were building an investment watchlist specifically around “expensive incumbent SaaS gets rebuilt from scratch by AI”, I'd rank the thesis roughly:

1. Rillet — strongest demonstrated ERP replacement signal
2. Harvey — strongest legal workflow attack
3. Sierra — strongest customer-service “outcome vs seat” thesis
4. Legora — formidable Harvey competitor
5. DualEntry — interesting ERP migration wedge
6. Campfire — another serious NetSuite challenger
7. Decagon — agent-native support
8. EvenUp — vertical legal workflow → outcome
9. Clay — GTM workflow reconstruction
10. Glean — enterprise knowledge/work orchestration

The common denominator isn't really AI.

It's removing the human-shaped architecture of SaaS.

Legacy SaaS asks:

“Which screens does the employee need to operate?” AI-native software asks:

“What outcome does the company want, and how much of the work can the system perform itself?” That shift also explains why seat-based pricing is vulnerable. An AI-native entrant can charge per transaction, outcome, resolution, filing, close, claim or completed workflow. Incumbents often can't do that without cannibalizing their own seat revenue.

And there is now evidence this isn't just an investor narrative: companies are actually ripping out incumbent systems. Rillet says roughly half of its customers migrated from legacy ERP/accounting products, with customers specifically replacing Intuit, NetSuite, Sage Intacct and Oracle.

The tech landscape is undergoing a massive shift from traditional "system of record" SaaS—which charges per human seat to help people log data and push buttons —to AI-native "system of action" (or Agentic) software . Instead of giving a human a tool, these companies build AI agents that do the work itself and charge based on usage, output, or labor displacement rather than per seat.

While legacy SaaS providers (like Salesforce, HubSpot, and Zendesk) are frantically adding AI sidebars and assistant features, a new wave of challengers is rebuilding workflows from scratch.

1. Customer Support & Service Operations

• The Incumbent Model: Zendesk, Intercom, Salesforce Service Cloud (priced per support agent seat, heavily reliant on human-driven ticket routing and macro responses).
• The AI-First Rebuild: Companies like Sierra (co-founded by Bret Taylor and Clay Bavor) and specialized autonomous support platforms don't just help support reps type faster—they sit at the front line, resolving upwards of 70–80% of tier-1 and tier-2 customer issues end-to-end across multiple backend enterprise systems. They monetize on successful resolutions rather than software licenses.

2. Legal, Compliance, & Contract Drafting

• The Incumbent Model: Complex contract lifecycle management (CLM) tools like Ironclad or legacy document management software where lawyers spend hours manually drafting, redlining, and reviewing clauses.
• The AI-First Rebuild: Legal tech and vertical agent startups (such as Harvey , which is backed heavily for legal workflows) are bypassing the traditional database/document repo model. They build the interface entirely around the generative legal reasoning loop—drafting, analyzing risk, and executing workflows based on raw context, effectively pricing per contract or outcome rather than charging a firm per paralegal seat.

3. Sales Prospecting, GTM, & Outbound CRM

• The Incumbent Model: ZoomInfo for data + Outreach/Salesloft for sequencing + Salesforce for tracking. It requires an army of SDRs (Sales Development Representatives) to manually filter and personalize emails.
• The AI-First Rebuild: AI-native platforms like 11x (building AI "digital workers" like Alice, an autonomous SDR) and Clay (which completely re-architected outbound data enrichment and waterfall enrichment as a programmable workflow) are replacing the rigid structure of traditional CRMs for top-of-funnel work. Instead of buying 20 software seats for an outbound team, companies deploy agents that research, write, and execute outreach autonomously.

4. Accounting, Finance, & Spend Management

• The Incumbent Model: Expensify, Concur, or NetSuite—software where employees log receipts and accountants manually chase approvals and code invoices.
• The AI-First Rebuild: AI-native spend and bookkeeping platforms are replacing manual invoice matching and expense policy enforcement with ambient background agents. They ingest unstructured receipts, cross-reference company policy, talk to vendors, and push verified journal entries directly to the ledger with near-zero human intervention.

Key Traits of These AI-First Challengers

• Outcome/Usage-Based Pricing: They are aggressively killing per-seat pricing . If an AI agent does the work of three data entry clerks or legal associates, the software charges a fraction of that labor cost or per completed task.
• No Database Forms to Fill Out: Traditional SaaS forced humans to populate fields so the software had data. AI-native workflows ingest unstructured data (emails, PDFs, voice, chat) and self-organize.
• The "Draft-and-Approve" Loop: They build trust iteratively . The agent starts by doing 90% of the work and presenting a clean approval dashboard to a human, graduating to full autonomy as the business gains confidence.

Are you looking at this shift from an investment/market perspective (evaluating the public SaaS drawdown) or looking to implement AI-native workflows in a specific department like sales, finance, or customer operations?

Yes. The interesting 2026 cohort isn't really “AI features for SaaS.” It's AI-native companies asking whether the incumbent workflow should exist at all.

That distinction matters: an AI copilot bolted onto Salesforce is not attacking Salesforce. An agent that owns prospecting → qualification → CRM → outreach → follow-up is.

A useful way to map the landscape:

The ones I'd watch most closely

1. Monaco — probably the cleanest “CRM from scratch” example.

Monaco launched in February with $35M raised and explicitly positioned itself against Salesforce/HubSpot. It has an AI-native CRM, a prospect database, and agents that research accounts, determine whom to contact, execute sequences and schedule meetings. The important part is that the CRM isn't the product's center of gravity—the agent executing the sales process is.

That's a much more dangerous architecture for Salesforce than “AI-powered CRM.”

2. Rillet / DualEntry / Campfire — the AI-native ERP cluster.

This is probably the strongest category right now. The thesis is essentially:

Don't make accountants operate NetSuite faster. Make the accounting system operate itself. Rillet, DualEntry, Campfire and Light are explicitly targeting NetSuite/Sage/QuickBooks-type systems with AI-native financial infrastructure. The category has attracted hundreds of millions of dollars in funding in a very short period.

DualEntry is particularly interesting because its “NextDay Migration” attacks one of the biggest incumbent moats: implementation and migration pain, not just product functionality. It claims to compress migrations that traditionally take months into roughly a day.

That's exactly the sort of thing that makes an incumbent vulnerable: AI isn't merely making the replacement better; it is destroying the switching-cost moat.

3. Harvey — professional software gets rebuilt around the worker.

Harvey is a slightly different beast because it isn't merely trying to be “AI Clio.” It's moving toward a system in which the AI actually performs substantial legal work.

Its latest platform incorporates persistent memory and domain-specific workflows, and the company is now building its own legal model. Harvey reportedly reached roughly $350M ARR and an $11B valuation this year.

This is the broader pattern: the database/application becomes the substrate; the agent becomes the worker.

4. Rox — revenue operations without the SaaS stack.

Rox is especially interesting because it attacks the stack, not a single incumbent. It connects to Salesforce, Zendesk and other systems but deploys hundreds of agents to research prospects, monitor accounts and update CRM data.

It's not quite as pure as Monaco because it still sits on top of existing infrastructure, but strategically it's pointing toward:

Salesforce + Gong + Clari + Outreach + ZoomInfo → one agentic revenue system.

That's an important pattern to watch.

5. Decagon — customer support without the traditional support application.

Customer support is arguably one of the best AI-native wedges because the ultimate unit of value isn't a ticket; it's a resolved customer problem.

Decagon is one of the more prominent startups pursuing that model, while Zendesk itself is being forced to respond by repositioning around an “Autonomous Service Workforce” and outcome-based pricing. Zendesk's own language is telling: it is moving from conventional bots toward specialized agents that actually resolve issues.

So this category may ultimately be a battle between AI-native challengers and incumbents forced to rebuild themselves.

---

The bigger opportunity isn't “AI SaaS”

I think the more interesting taxonomy is:

1. AI replaces the user interface

Old:

Employee → forms → workflows → database

New:

Employee → intent → agent → database/actions

This attacks Salesforce, Workday, ServiceNow, etc.

2. AI replaces the workflow

Old:

Lead → SDR → CRM → enrichment → sequencing → meeting → CRM update

New:

Goal → agent → outcome

Monaco/Rox are examples.

3. AI replaces the labor inside the workflow

Old:

Customer issue → ticket → support agent → escalation

New:

Customer issue → agent → resolution → human only when necessary

This is the Decagon-style opportunity.

4. AI destroys the implementation layer

This one might be the sleeper.

Historically:

Buy software → consultants → customize → migrate data → train employees → finally use software

AI-native:

Connect data → agent understands business → system configures itself

DualEntry's migration thesis is an example. If AI makes implementation 10× cheaper and faster, implementation cost stops protecting incumbents.

---

Where I think the biggest white spaces are

If you're looking for companies that could become the “Cursor of X”, I'd put the categories roughly here:

1. ERP / finance — already happening; strongest evidence.
2. Sales / CRM — very early, but potentially enormous.
3. Customer support — perhaps the most immediately automatable.
4. Legal — AI can directly perform expensive knowledge work.
5. Procurement — enormous amount of repetitive, document-heavy workflow.
6. Insurance operations — underwriting, claims, policy servicing.
7. Healthcare administration — prior auth, billing, scheduling, intake.
8. HR / recruiting — potentially an entirely different hiring operating system.
9. Compliance / risk — continuous agentic monitoring instead of periodic workflows.
10. IT operations — ServiceNow's ticket/workflow abstraction is particularly exposed.

The key isn't simply “lots of AI.” The strongest startups have a different unit of abstraction.

Instead of:

“Here's better software for the employee.” they're saying:

“Why does the employee need to use software for this at all?” That's the fundamental threat to expensive incumbent SaaS. And it's why the current software disruption looks more like workflow replacement than conventional feature competition. The broader SaaS market is already seeing the build-vs-buy boundary shift as coding agents make custom software dramatically cheaper.

If I were building an investment map, I'd put Monaco, Rillet, DualEntry, Campfire, Harvey, Decagon and Rox in the first “serious AI-native challengers” bucket, then look for the next wave attacking ServiceNow, Workday, Coupa, Okta, and vertical-specific systems of record.

Don't make accountants operate NetSuite faster. Make the accounting system operate itself. Rillet, DualEntry, Campfire and Light are explicitly targeting NetSuite/Sage/QuickBooks-type systems with AI-native financial infrastructure. The category has attracted hundreds of millions of dollars in funding in a very short period.

DualEntry is particularly interesting because its “NextDay Migration” attacks one of the biggest incumbent moats: implementation and migration pain, not just product functionality. It claims to compress migrations that traditionally take months into roughly a day.

That's exactly the sort of thing that makes an incumbent vulnerable: AI isn't merely making the replacement better; it is destroying the switching-cost moat.

3. Harvey — professional software gets rebuilt around the worker.

Harvey is a slightly different beast because it isn't merely trying to be “AI Clio.” It's moving toward a system in which the AI actually performs substantial legal work.

Its latest platform incorporates persistent memory and domain-specific workflows, and the company is now building its own legal model. Harvey reportedly reached roughly $350M ARR and an $11B valuation this year.

This is the broader pattern: the database/application becomes the substrate; the agent becomes the worker.

4. Rox — revenue operations without the SaaS stack.

Rox is especially interesting because it attacks the stack, not a single incumbent. It connects to Salesforce, Zendesk and other systems but deploys hundreds of agents to research prospects, monitor accounts and update CRM data.

It's not quite as pure as Monaco because it still sits on top of existing infrastructure, but strategically it's pointing toward:

Salesforce + Gong + Clari + Outreach + ZoomInfo → one agentic revenue system.

That's an important pattern to watch.

5. Decagon — customer support without the traditional support application.

Customer support is arguably one of the best AI-native wedges because the ultimate unit of value isn't a ticket; it's a resolved customer problem.

Decagon is one of the more prominent startups pursuing that model, while Zendesk itself is being forced to respond by repositioning around an “Autonomous Service Workforce” and outcome-based pricing. Zendesk's own language is telling: it is moving from conventional bots toward specialized agents that actually resolve issues.

So this category may ultimately be a battle between AI-native challengers and incumbents forced to rebuild themselves.

---

The bigger opportunity isn't “AI SaaS”

I think the more interesting taxonomy is:

1. AI replaces the user interface

Old:

Employee → forms → workflows → database

New:

Employee → intent → agent → database/actions

This attacks Salesforce, Workday, ServiceNow, etc.

2. AI replaces the workflow

Old:

Lead → SDR → CRM → enrichment → sequencing → meeting → CRM update

New:

Goal → agent → outcome

Monaco/Rox are examples.

3. AI replaces the labor inside the workflow

Old:

Customer issue → ticket → support agent → escalation

New:

Customer issue → agent → resolution → human only when necessary

This is the Decagon-style opportunity.

4. AI destroys the implementation layer

This one might be the sleeper.

Historically:

Buy software → consultants → customize → migrate data → train employees → finally use software

AI-native:

Connect data → agent understands business → system configures itself

DualEntry's migration thesis is an example. If AI makes implementation 10× cheaper and faster, implementation cost stops protecting incumbents.

---

Where I think the biggest white spaces are

The market is moving past the era where SaaS is just a database with a human-operated interface [1.22]. Instead of traditional software that waits for a user to click buttons, fill out fields, and push data across tabs, AI-native platforms are building autonomous, outcome-driven workflows.

A wave of well-funded AI companies is actively bypassing or displacing expensive incumbent SaaS by redesigning entire operational stacks from scratch.

1. Go-To-Market (GTM) & Sales Data

• Incumbents being attacked: ZoomInfo, Apollo, legacy data brokers, and manual prospecting tools.
• The AI-native disruptor:
• How they rebuilt the workflow: Instead of buying static contact databases, Clay treats the entire internet as a live database. It introduced a "waterfall enrichment" framework that scrapes data from over 150+ providers, runs automated web searches, and applies custom AI logic to qualify prospects on the fly. Rather than paying for hundreds of bloated human-seat subscriptions to do manual list building, revenue teams use Clay to orchestrate multi-step data pipelines and trigger outreach programmatically.

2. Customer Support & Service Operations

• Incumbents being attacked: Zendesk, Salesforce Service Cloud, Intercom's legacy tiers.
• The AI-native disruptors: Sierra AI and Decagon
• How they rebuilt the workflow: Traditional helpdesks were built around human agents sorting tickets. AI-native support layers are built around "containment and resolution"—meaning the AI doesn't just draft a reply; it authenticates users, queries backend ERPs/databases, handles multi-turn troubleshooting, and executes refunds autonomously. Rather than pricing per human seat (which penalizes the software as AI reduces headcount), these platforms lean heavily into resolution-based or outcome-based economics.

3. Healthcare Clinical Documentation

• Incumbents being attacked: Epic, Cerner manual clerical entry, and traditional medical transcription services.
• The AI-native disruptor:
• How they rebuilt the workflow: Doctors used to spend two hours on clerical data entry for every one hour of direct patient care. Abridge utilizes ambient AI listening to securely capture the actual natural conversation between a clinician and a patient, instantly extracting medical facts and structuring them into billable notes directly inside the Electronic Health Record (EHR). It shifts the doctor’s job from manual typing to a quick review-and-approve workflow, dramatically reducing cognitive load.

4. Legal Research & Contract Intelligence

• Incumbents being attacked: Thomson Reuters, LegalZoom, legacy document management systems.
• The AI-native disruptor: Harvey
• How they rebuilt the workflow: Traditional legal tech provided databases for lawyers to search manually. Harvey structures the legal workflow around the "matter" or project itself. With its next-generation architecture (such as Harvey II's persistent memory and domain-tuned models like Tenet), the platform handles multi-jurisdictional research, automated contract markups, and contextual drafting in a firm's precise voice—acting as an autonomous accelerator for complex legal tasks rather than a passive form-filler.

Would you like to explore how these AI-native platforms are shifting their pricing models away from per-seat subscriptions toward consumption and outcome-based billing?

Yes. If by “attacking expensive incumbent SaaS” you mean companies that aren't merely adding a copilot, but are asking “what would this workflow look like if an AI agent—not a human clicking through forms—were the primary user?”, there is a pretty interesting cohort emerging.

The strongest examples I’d watch in 2026:

The ERP category is probably the cleanest proof

This is where the thesis is easiest to see.

Traditional ERP is essentially:

database → forms → rules → human operator → reports

The new architecture is closer to:

business event → AI agent → actions across systems → continuously reconciled state

Rillet, DualEntry, Campfire and others are explicitly positioning themselves against NetSuite/Sage/QuickBooks rather than merely offering an AI assistant for them. An industry review of the 2026 market identifies Rillet, DualEntry, Campfire and Light as the leading AI-native ERP cohort, with the products targeting NetSuite and other incumbent suites.

DualEntry is particularly interesting because the migration itself is part of the attack. Its “NextDay Migration” uses AI to move historical financial data from a legacy ERP in roughly 24 hours rather than the months traditionally associated with ERP migrations. Reuters reported that the company raised $90M at a $415M valuation specifically around this ERP-disruption thesis.

And qomplement is even more explicit: it says it is building an agentic ERP that replaces NetSuite/SAP and has agents handle procurement, inventory, freight and finance end-to-end.

That's much more interesting to me than “NetSuite, but with an AI chatbot.”

Sierra is the other archetype

Customer service is another unusually good market for this strategy.

The old stack is:

Zendesk/Salesforce → ticket → human reads ticket → searches knowledge base → clicks through systems → responds

The AI-native version is:

customer request → agent understands intent → agent investigates → agent takes actions → agent resolves → human only handles exceptions

That's why I put Sierra in the same conceptual bucket as the AI-native ERP companies, even though the product category is completely different. Its acquisition of Takeoff in July 2026 is also telling: Sierra described the combined direction as a platform for long-horizon agents, rather than simply an AI interface on top of conventional software.

Glean is slightly different

I wouldn't put Glean in the pure “kill SaaS” bucket.

It's building an AI control/intelligence layer across SaaS, connecting systems such as Slack, Google Drive, Jira and Salesforce and allowing agents to take action. Glean itself now describes the product as connecting knowledge, systems and context so AI can actually perform work.

Its $300M ARR milestone in May 2026 is evidence that this “layer above the SaaS stack” strategy is working commercially.

So I think there are actually three strategies emerging:

1. Replace the system of record
   Rillet, DualEntry, Campfire, qomplement, Puzzle

2. Replace the human workflow sitting on top of the system of record
   Sierra, Harvey and similar agentic applications

3. Become the AI operating layer across the existing SaaS stack
   Glean

And #1 is the really provocative one.

---

The companies I'd watch most closely

If I were building an investment/competitive landscape rather than just making a long startup list, I'd put five names in the first row:

Rillet → DualEntry → Sierra → Harvey → qomplement

because they're attacking fundamentally expensive software/workflows rather than selling another productivity layer.

The underlying bet is:

AI doesn't necessarily make SaaS better. It may make the assumptions behind SaaS unnecessary.

That's a much bigger thesis.

Traditional SaaS assumes a human needs a specialized application with screens, permissions, workflows, dashboards and hundreds of configuration options.

An agent doesn't necessarily need any of that.

If the agent can understand the business context, query the underlying data, execute transactions and maintain an auditable state, the application itself can collapse into an AI-driven workflow.

That's also why the current SaaS shakeout is different from ordinary feature competition. Recent reporting has explicitly framed the pressure as AI-native startups attacking workflows that legacy SaaS was built around, rather than merely competing on features.

My highest-conviction category is AI-native ERP/accounting. It's unusually attractive because incumbent software is expensive, implementation is painful, workflows are repetitive, and the underlying data is structured enough for agents to operate on it. The market is already attracting substantial capital.

If you want, I can also give you a much more interesting list of ~30 startups, organized by incumbent (Salesforce, Workday, ServiceNow, SAP, NetSuite, Atlassian, HubSpot, etc.), and rank each one by how credible the “kill the incumbent” thesis actually is.

The software landscape is undergoing a massive structural shift away from traditional, dashboard-heavy, per-seat SaaS and toward AI-native, agent-driven execution models . Instead of giving a human worker a tool to organize data (like a ticketing queue or a CRM form), these new challengers build systems where autonomous AI executes the workflow end-to-end and charges based on outcomes or usage rather than human licenses.

The most prominent categories and the AI-native companies disrupting them from scratch include:

• Customer Service & Support (Attacking Zendesk, Salesforce Service Cloud)
  • Sierra : Co-founded by Bret Taylor, Sierra scales zero-to-one customer interaction by focusing purely on autonomous resolution and pricing strictly on outcomes rather than per-seat licenses.
  • Decagon & Ada : These platforms build deep generative support workflows that ingest product logs, process refunds, and resolve complex edge cases without routing tickets to human operators.
  • Intercom (Fin) : While an incumbent, Intercom aggressively transitioned its core product philosophy around Fin to handle automated Tier 1-3 resolution natively, setting an outcome-based standard that forces legacy pure-play helpdesks to adapt or compress.
• Software Development & Engineering (Attacking Traditional Dev-Ops and IDE Tooling)
  • Cursor : Rebuilt the coding workflow from scratch as an AI-first IDE rather than just a plugin, capturing massive developer mindshare and forcing legacy tools to scramble.
  • Lovable / Vibe-Coding Ecosystems : These platforms allow users to generate functional application workflows and full-stack execution using natural language prompt layers, bypassing traditional rigid tool stacks.
• Sales, CRM, & Revenue Operations (Attacking Salesforce, HubSpot)
  • AI-First SDR and Revenue Startups : A new wave of YC-backed and independent agentic platforms (like emerging vertical CRM replacement layers) are taking over lead enrichment, signal capture, automated multi-channel sequencing, and pipeline updates. Instead of reps logging into Salesforce dashboards 20 times a day, agents autonomously listen to buying signals and execute CRM hygiene.
• IT Service Management & Internal Operations (Attacking ServiceNow, Jira Service Management)
  • Console & Atomicwork : These AI-native ITSM platforms are built on the assumption that internal IT, HR, and finance workflows can be orchestrated via natural language in Slack/Teams without a human manually categorizing or routing requests.

The Core Dynamic: Features vs. Workflows

Incumbent SaaS giants (like Salesforce with Agentforce , ServiceNow with Now Assist , and Workday) are aggressively trying to retrofit intelligence into their massive legacy databases . However, AI-native challengers are winning by owning the action layer and pricing on outcomes (e.g., per contract drafted, per resolution achieved, or percentage of recovery) rather than charging you for empty software licenses while your human headcount shrinks.

If you want to narrow this down, let me know:

• Which specific SaaS category (e.g., HR, Finance, Legal, Marketing) you are most interested in tracking
• Whether you want to focus on venture-backed startups or how incumbents are fighting back

I can dive deeper into the economics and specific players for that space.

The transition from traditional, human-centric Software-as-a-Service (SaaS) to AI-native, agentic workflows is fundamentally rewriting enterprise software. Instead of legacy SaaS models—which digitized manual processes with form-filled dashboards and charged per human "seat"—these challengers rebuild the workflow from the ground up so that AI agents execute the work itself.

Startups and scale-ups across healthcare, legal, customer operations, and revenue systems are leading this shift:

• Abridge & Freed (Healthcare Scribe/EHR Workflows): Instead of navigating bloated Electronic Health Record (EHR) menus or paying massive administrative overhead for human medical transcription, these platforms ingest raw, unstructured ambient audio from doctor-patient conversations and autonomously structure, code, and log clinical notes directly into legacy systems like Epic.
• Intercom (Customer Operations via Fin): While historically a traditional support ticketing SaaS, Intercom effectively refounded its product around Fin , an AI agent capable of resolving millions of end-to-end customer support requests autonomously rather than routing them through human agents sitting at software "seats".
• Prepared (Public Safety & 911 Workflows): Prepared modernizes emergency dispatch by ingesting unstructured live video and text data from 911 callers, then using autonomous voice and data agents to transcribe, summarize, and log information directly into rigid Computer-Aided Dispatch (CAD) systems of record.
• Hyperscience (Government & Insurance Data Processing): Rather than traditional data entry or legacy OCR software that requires heavy human verification, Hyperscience built a software core designed around machine-learning interpretation and automation to process unstructured documents end-to-end.
• Outreach & RevOps Challengers (Revenue Operations): Traditional sales engagement platforms are being pressured by agent-driven systems that autonomously update CRMs, conduct multi-channel prospecting, and handle the execution layer of revenue operations rather than just giving reps a dashboard to do it manually.

The Shift in How the Workflow is Rebuilt

1. From Systems of Record to Systems of Action: Legacy SaaS merely stored data for humans to look at and organize. AI-native alternatives act on the data directly, replacing the human labor cycle rather than assisting it.
2. From Seat-Based to Outcome-Based Pricing: Because AI agents do the work of entire teams in milliseconds, per-seat subscription models are being replaced by usage- or outcome-based pricing (charging per contract drafted, ticket resolved, or claim processed).

If you'd like to narrow this down, let me know which specific industry or department (e.g., legal, finance, HR, customer support) you are tracking, and I can give you a deeper breakdown of the AI-native challengers operating there.

Yes. The interesting wave is not “AI features for SaaS.” It’s companies asking: if agents can actually do the work, why do we need the old application at all?

A useful way to map the landscape in 2026:

The ones I'd watch most closely

1. Sierra — customer support

This is probably the cleanest example of the thesis. Instead of selling another support-agent interface, Sierra is effectively selling resolution of the customer's problem. The economic unit starts moving from seats/tickets to outcomes. Sierra reportedly crossed $100M ARR and was valued around $10B in early 2026.

2. Decagon — customer support

Similar attack, but with an especially strong replacement thesis: Decagon customers have reportedly replaced existing IVR, ticketing, or CRM-based agent systems rather than simply adding Decagon alongside them.

This is important because it demonstrates the difference between:

“AI makes Zendesk better”

and

“You don't need Zendesk for this workflow anymore.”

3. Rillet / DualEntry / Campfire — ERP

This might be the most consequential category.

Traditional ERP is an unusually attractive target because implementation is painful, interfaces are designed around humans entering structured data, and companies spend enormous amounts on consultants and administrators.

• Rillet → AI-native finance/ERP, attacking NetSuite/Sage/QuickBooks.
• DualEntry → particularly interesting because its AI-assisted migration attacks one of ERP's biggest moats: switching costs.
• Campfire → broader finance operating system, with startups actually moving off NetSuite.
• Light → global/multi-entity finance.
• qomplement → much more aggressive: explicitly pitches itself as an agentic ERP replacing SAP/NetSuite in supply-chain operations.

The ERP landscape is arguably the clearest instance of “rebuild the system around the agent rather than bolt an agent onto the system.”

4. Harvey — legal

Harvey is a slightly different version of the same idea: don't build “AI inside legal software”; build the product around getting legal work completed.

That means the unit of value increasingly becomes research, drafting, diligence, review, etc., rather than licenses for lawyers. BCG's 2026 analysis puts Harvey among the AI-native startups already reaching meaningful enterprise AI revenue.

5. Monaco / Aurasell / Lightfield / Rox — CRM & GTM

This is probably the most interesting battlefield if your definition is specifically “attack a giant horizontal SaaS incumbent.”

The old stack can look like:

Salesforce + ZoomInfo + Apollo + Outreach + Gong + Clari + CPQ + spreadsheets + sales ops people

The AI-native thesis is:

one system + agents → research → prospecting → outreach → calls → CRM → forecasting → pipeline management

Aurasell is unusually explicit about this consolidation thesis, claiming it can replace 15+ GTM tools, while also providing a transitional layer on top of Salesforce/HubSpot.

Monaco is taking a more radical “AI sales team” approach, while Lightfield is attacking the CRM itself. The broader AI-GTM field now includes Rox, Monaco, Aurasell, Artisan, Landbase and others.

---

The pattern I'd actually pay attention to

There are three levels of AI SaaS, and only the third is really threatening incumbents:

Level 1 — AI feature

Salesforce + AI
Zendesk + AI
Workday + AI

Incumbent wins.

Level 2 — AI copilot

“Here's an agent that helps you use Salesforce.”

Better, but the incumbent still owns the system of record.

Level 3 — AI-native workflow

“Why are you using Salesforce at all? Tell us what outcome you want and we'll execute the workflow.”

That's the interesting one.

The architectural change is profound: the database/UI stops being the product and becomes infrastructure. The agent becomes the primary interface, and the company sells the work performed rather than access to software.

That's why Ramp's 2026 spending data is interesting: AI-native vendors are already gaining adoption share in categories such as sales execution, BI and product analytics even though incumbents still capture most of the dollars.

And it explains why the incumbents are responding with Agentforce, ServiceNow AI Agents, Workday's Sana, SAP Joule, etc. They're trying to transform themselves from applications humans operate into systems agents operate.

My shortlist of the most interesting attacks

If I were looking for companies with the strongest “throw away the old SaaS” thesis rather than merely “AI-enabled SaaS,” I'd put these near the top:

Sierra → Zendesk / customer-support stack
Decagon → support stack
Rillet → NetSuite
DualEntry → NetSuite / accounting
qomplement → SAP / NetSuite for supply chain
Harvey → legal workflow
Monaco → sales org / GTM stack
Aurasell → Salesforce + GTM stack
Lightfield → CRM

The common denominator is not LLM quality. It's that they are willing to redesign the workflow, data model, UI, pricing and labor model simultaneously. That's much harder for an incumbent to do without cannibalizing its existing business.

And there's a fascinating counterpoint: Salesforce, ServiceNow and Workday are increasingly trying to make their existing systems agent-native rather than human-native, so the ultimate contest may be AI-native startup vs. incumbent with 20 years of proprietary workflow/data/integration infrastructure.

If you're thinking about this from a startup-investing angle, I can also rank ~30 companies by “probability of actually killing a $1B+ incumbent SaaS category” rather than just listing AI-native startups.