What the SaaS stack does well
The honest answer is: a lot of it is good, up to a point. Stripe is excellent. It handles payment processing, fraud detection, subscription management, and checkout with a reliability that would take years to replicate from scratch. Klaviyo's email infrastructure is stable, deliverable, and continuously improved. Rebuy applies collaborative filtering at scale without requiring a data science team.
The SaaS stack earns its place for stores that are early-stage, growing quickly, and don't need non-standard behavior. The comparison to custom agents only becomes relevant when the specific store's needs exceed what standardized, maintained, continuously improved infrastructure can deliver — or when the store needs infrastructure that SaaS tools structurally cannot provide.
What the SaaS stack structurally cannot do
It cannot make your store ACP-compliant. The Agentic Commerce
Protocol (ACP) requires a discovery file at /.well-known/agent-purchase.json
and a purchase endpoint your backend exposes. No SaaS commerce tool deploys this
automatically. ChatGPT, Operator, and Perplexity shopping agents follow ACP to
route purchases. Stores built on SaaS infrastructure will need to build ACP
compliance on top of tools they do not control.
It does not share context across tools. The cart abandonment email Klaviyo sends does not know what the Rebuy recommendation widget showed before the cart was abandoned. The post-purchase sequence in Aftersell does not know the customer's session history before the sale. Each tool operates in its own data silo. The actions are individually optimized but collectively uncoordinated.
It does not adapt to your specific store's behavior patterns. Klaviyo's abandonment flow is built for the average cart abandonment trigger. Your abandonment pattern may be driven by a specific shipping rate display, a specific product configuration page, or a specific device/browser combination that your analytics have surfaced but that the tool cannot act on without significant custom configuration — if it can act on it at all.
It does not get cheaper as you scale. Klaviyo's pricing is contact-based. More customers means more cost. Rebuy's pricing is revenue-based. More sales means more fees. The SaaS stack is designed to capture value as the store grows — which is good for the vendors and less good for the owner.
It does not produce owned infrastructure. After three years of paying for these tools, you have three years of access history and zero owned assets. The moment you stop paying, the capability disappears.
The real distinction is not APIs — it is who controls the orchestration
Agentic commerce still uses third-party APIs. Stripe handles checkout. Resend delivers emails. These integrations are not the problem — they are good infrastructure for specific, well-defined jobs. The problem is when the tool itself makes the decisions.
When Klaviyo decides the timing, the segmentation, and the abandonment logic, you are configuring fields inside their interface. The behavior lives in their system. When you need something they have not built, you wait for them to ship it. When they raise prices, you pay or you rebuild. When a competitor moves faster because their stack does something yours cannot, you are at the mercy of a roadmap you do not control.
Custom agents call Stripe's API on your terms. They call Resend's API with your logic, your timing, your conditions. The API key is in your environment file. The code is in your repository. You can change the provider, rewrite the logic, or extend the behavior without asking anyone's permission — and without waiting for a release cycle that serves ten thousand other stores before it serves yours.
This is where convenience costs you the front of the market. Every store using Klaviyo gets the same abandonment logic. Every store using Rebuy gets the same recommendation model. The standardization that makes SaaS tools easy to adopt is the same standardization that makes them impossible to differentiate on.
What custom agents add to the SaaS layer
The interesting strategic question is not "SaaS vs. custom" as a binary — it is where custom agents add value on top of what SaaS does correctly.
Stripe stays. Payment processing is infrastructure. Rebuilding it is unnecessary risk. Stripe's fraud detection, regulatory compliance, and payment method coverage are genuinely valuable.
The behavioral layer gets replaced. The agents that use session data, purchase history, and catalog-specific logic to make per-customer decisions — these are the candidates for replacement with custom agents. Not because the SaaS tools are bad, but because custom agents can be trained on your specific behavior patterns, share context across the lifecycle, and run at fixed cost as volume scales.
The monitoring layer gets added. None of the SaaS tools provide a unified view of all agent activity across the commerce lifecycle. JARVIS is purpose-built for that — to show what every agent is doing, what signals they are acting on, and what the outcomes are. That visibility does not exist in the fragmented SaaS stack.
The honest comparison by use case
Cart recovery: Klaviyo's default abandonment flow vs. a purpose-built cart recovery agent. The SaaS tool sends a generic three-email sequence. The agent identifies the trigger (price friction, shipping cost, product confusion, interrupted session) and responds to that signal. For a store with a $300 average order value and 200 abandoned carts per month, improving recovery rate by 5 percentage points is $30,000 in annual revenue.
Product discovery: Rebuy or LimeSpot vs. a purpose-built discovery agent. The SaaS tools apply collaborative filtering (customers who bought X also bought Y). The agent applies catalog-specific logic combined with session context. For stores with complex catalogs or unusual purchase patterns, the SaaS approach produces mediocre recommendations; the agent produces relevant ones.
Post-purchase: Aftersell or a Klaviyo post-purchase flow vs. a purpose-built post-purchase agent. The SaaS tools run the same sequence for every customer. The agent runs different sequences based on what the customer bought, how they bought it, and what their session history suggests about their next need.
SaaS tools build for the median customer at the median store. Your store is not the median. The gap between what a SaaS tool delivers and what a purpose-built agent delivers is exactly the size of the distance between your store and the median.
When the SaaS stack is still the right answer
Below $10,000/month in revenue, the SaaS stack is probably still the right answer. The data volume is too low to train contextually meaningful agents, the improvement delta is smaller, and the operational overhead of maintaining custom infrastructure is disproportionate.
For stores between $10,000 and $50,000/month, the Foundation build — one agent for one specific problem — is the right entry point. One well-built cart recovery agent or one product discovery agent can pay for the entire build within 90 days while keeping the rest of the SaaS stack in place.
For stores above $50,000/month with specific, known friction points across multiple stages of the commerce lifecycle, the Full Suite is the right evaluation. The SaaS costs at that revenue level are typically $2,000–$5,000/month, and the efficiency gains from contextually coordinated agents are larger.
What the SaaS stack does not replace: owned organic content
There is a second category of SaaS spending the build also addresses: SEO and content tools. Stores paying for Ahrefs, Semrush, or content agencies to generate organic traffic are running a subscription against a channel they do not own. JARVIS — included in the Both Together tier — replaces that with an autonomous content pipeline on your infrastructure. The articles live in your repository. The traffic they generate is yours permanently.
The cost of a mid-tier SEO tool runs $100–$500/month. A content agency producing consistent output runs $1,000–$3,000/month. JARVIS runs at approximately $0.001 per article in inference cost — fractions of a cent per piece of content that indexes and earns traffic indefinitely.
BEYOND THE SAAS STACK
Purpose-built agents,
zero SaaS dependency.
Foundation $5,000 — you only pay after a successful test. Full Suite $20,000 · Both Together $25,000 · Full source ownership.
Apply for a build → ← Back to Agentic Commerce overview