ICYMI: "We Don't Trust… | Sourcery Newsletter Summary

1. Key Themes

Theme 1: AI Agent Governance Is the Next Enterprise Infrastructure Layer

The single biggest emerging need in enterprise AI is not building agents — it's controlling them. Merge's Agent Handler product positions a central hub between employees' AI tools and downstream enterprise systems, enabling governed connectivity at scale.

"We can't let you just connect all of our internal services to all of your external services." (CTO of a large financial firm, as quoted by Gil)

"The only tool you're allowed to connect it to is Merge Agent Handler." — Gil Feig

This is a genuine market gap: employee demand for agentic automation exists, but regulatory and security constraints block it without a governance layer in between.

Theme 2: The Real AI Security Threat Is Access, Not Intelligence

Merge's security philosophy centers on a crucial distinction: an agent's danger is not its reasoning capability but what systems it can reach. This reframes the entire AI security conversation away from alignment and toward access control.

"An evil genius who's a mass murderer, but they're locked in a jail cell... The second you connect it to tools, which is what everyone is trying to do right now, that's where everything goes wrong." — Gil Feig

"We don't trust agents. We can try to set rules, but there need to be hard guardrails and blocks for things like sensitive data being sent across." — Gil Feig

Theme 3: AI Is Democratizing the Attacker, Accelerating the Threat Surface

The threat landscape has fundamentally shifted because AI removes the skill barriers that previously limited bad actors. This is not theoretical — Merge observed it in production.

"Merge saw over 1,000 bot signups in an hour, scanning backend endpoints from IPs around the world."

Attackers who once faced language and skill barriers now "speak English perfectly," "write code perfectly," and "have unlimited manpower all driven by AI." — Gil Feig

Supply chain attacks are compounding this: "Agents are pushing a ton of code. You don't have enough humans to read all that code, and so things are slipping by." — Gil Feig

Theme 4: AI Infrastructure Costs Are Creating a CFO Reckoning

The "token-maxxing" enthusiasm of AI teams is running headlong into budget reality, creating a new enterprise pain point around AI cost optimization and model routing.

"The bill comes to the CFO, and it's actually really brutal and way worse than they expected." — Gil Feig

"There are a lot of really great open source providers out there too that are very cheap... routing a query like 'what's one plus one' or a simple 'thanks' to a cheaper model can cut spend, but it's really hard to automate that." — Shensi Ding

This dynamic is creating real demand for model gateway infrastructure — a product category with structural tailwinds as AI usage scales.

Theme 5: The SaaSpocalypse — Traditional SaaS Is Under Structural Pressure

The cost of building in-house software has collapsed, removing one of enterprise SaaS's core moats: switching costs justified by build complexity.

"The time to build the same exact product in-house is just significantly lower." — Shensi Ding

"The old negotiating line of 'I could just build this' used to carry a real cost. Today that cost is 'very, very cheap.'"

"It gets to the point where English becomes your programming language." — Gil Feig

2. Contrarian Perspectives

Perspective 1: Companies Over-Engineering Their Own Models Is a Strategic Mistake

While the market celebrates vertical AI model development, Shensi offers a pointed counter-view: most companies should not be building their own models. This is a hot take that pushes against a significant amount of current enterprise AI investment.

Shensi shares "a hot take on companies over-engineering their own models."

The implication: the real leverage is in connectivity, governance, and routing — not in fine-tuning or training — and companies burning resources on custom models may be solving the wrong problem.

Perspective 2: AI Valuations May Actually Make Sense — But Not for the Reasons People Think

Rather than dismissing sky-high AI valuations, the founders engage with the possibility that the market may be rationally pricing in a structural shift. The episode raises the question — "Are AI valuations actually insane?" — without a dismissive answer, suggesting the founders see legitimate underlying value that critics miss.

Supporting context: Merge itself is now powering infrastructure for OpenAI, JPMorgan, Netflix, Perplexity, Uber, Mistral, and Dropbox — a customer list that validates the scale of AI enterprise adoption. The market may be pricing in winner-take-most infrastructure dynamics that traditional SaaS multiples don't capture.

Perspective 3: Don't Count Out Marc Benioff and Legacy Giants

Against the conventional narrative that AI-native startups will displace legacy SaaS incumbents, Shensi expresses genuine respect for Benioff's ability to navigate technological transitions — pointing to the concept of the "beginner's mind" as his durable edge.

"I just would not count Benioff out." — Shensi Ding

Shensi admires Benioff's "beginner's mind" approach and notes how difficult it is to "sustain a dominant product through multiple tech shifts."

The move to "headless Salesforce" — where agents interact via API or CLI without a UI — is cited as evidence that incumbents are adapting faster than skeptics expect.

3. Companies Identified

Merge

Description: AI infrastructure company offering a three-product suite: Unified (data sync/normalization), Agent Handler (governed agentic connectivity), and Gateway (AI model routing and cost optimization)
Why Mentioned: Primary subject of the article; raised $75M from Accel, Addition, and NEA
Quote: "The leading provider of customer-facing integrations and agentic tools for frontier LLMs, Fortune 500 organizations, and B2B SaaS companies."

OpenAI

Description: Leading AI lab
Why Mentioned: Named as a Merge customer; cited as an example of AI companies with fast, low-friction sales cycles
Quote: "How Merge landed OpenAI, Perplexity, Netflix & Uber" (timestamp reference)

JPMorgan

Description: Global financial services giant
Why Mentioned: Named as a Merge customer, representative of large financial services firms that begin with Unified API for deterministic use cases

Perplexity

Description: AI-native search company
Why Mentioned: Named as a Merge customer; example of AI-native buyers with fast deal cycles

Netflix

Description: Global streaming platform
Why Mentioned: Named as a Merge customer; represents large-scale enterprise adoption of Merge's infrastructure

Uber

Description: Ride-sharing and logistics platform
Why Mentioned: Named as a Merge customer

Mistral

Description: European AI model company
Why Mentioned: Named as a Merge customer

Dropbox

Description: Cloud storage and collaboration platform
Why Mentioned: Named as a Merge customer

Salesforce

Description: Enterprise CRM platform
Why Mentioned: Used as a case study for the "headless software" shift — agents interacting with enterprise software via API/CLI rather than UI
Quote: "What does Salesforce going headless exactly mean" (timestamp); "I just would not count Benioff out." — Shensi Ding

Cursor

Description: AI-powered code editor
Why Mentioned: Cited as a signal of improving AI margin discipline in the industry
Quote: Shensi cited "Cursor's margin improvements as a signal of where things are heading."

Wiz

Description: Cloud security company
Why Mentioned: Referenced in the context of the GitHub hack and supply chain security discussion (timestamp: "Mythos, Wiz, and the GitHub Hack")

4. People Identified

Shensi Ding

Description: CEO and co-founder of Merge
Why Mentioned: Primary interview subject; shares strategic vision, hiring philosophy, and perspective on the SaaSpocalypse and AI market
Quote: "The best way to succeed is to just do things. If you over-intellectualize your company building instead of actually doing anything, you're too high on Maslow's hierarchy."

Gil Feig

Description: CTO and co-founder of Merge
Why Mentioned: Primary interview subject; drives the technical and security architecture perspective at Merge
Quote: "If we are the leaders of this company, we have to know everything there is to know about how AI works, how you build with AI."

Marc Benioff

Description: CEO of Salesforce
Why Mentioned: Cited by Shensi as an example of a founder with a "beginner's mind" who should not be underestimated through multiple tech transitions
Quote: "I just would not count Benioff out." — Shensi Ding

Keith Rabois

Description: Venture capitalist and operator (Founders Fund, previously PayPal, Square)
Why Mentioned: Referenced for his framework of "barrels" vs. "ammunition" as a talent filter Merge uses in hiring
Quote: Merge looks for Keith Rabois' "barrels" who push to get things done over more task-specific "ammunition."

5. Operating Insights

Insight 1: Founder-Led AI Literacy Is Non-Negotiable

Merge's pivot succeeded because both founders personally went back to building with AI tools — not by delegating the learning. This gave them firsthand insight into what their own products could do in an AI context, which unlocked the next product generation.

"If we are the leaders of this company, we have to know everything there is to know about how AI works, how you build with AI." — Gil Feig

"That month actually really transformed the business" — referring to the window when a stalled deal forced them to build hands-on with AI coding tools.

Takeaway for operators: Don't outsource your AI education to a team. Personal fluency changes what you're able to see in your own product.

Insight 2: Hire Missionaries, Filter Out Mercenaries — Especially in a Hype Cycle

With AI valuations creating a mercenary talent market, Merge explicitly filters candidates based on mission alignment versus compensation-chasing.

They hire for "missionaries versus mercenaries," filtering out candidates chasing flashy valuations.

They look for Keith Rabois' "barrels" who push to get things done over more task-specific "ammunition."

Takeaway for operators: In a frothy market, the signal-to-noise ratio in hiring degrades. Explicit cultural filtering for motivation type is a high-leverage operating decision.

Insight 3: Be Prescriptive with AI Customers Who Don't Know What They Want

AI buyers in 2024-2025 often lack the technical depth of prior SaaS buyers and are making purchasing decisions in a category that is still being defined. The winning sales motion is consultative, not reactive.

"The team draws on its experience to guide customers toward best practices, since many 'don't have experience with partnerships for these different integrations' and aren't sure what the best end-user experience looks like."

The product has had to adapt to "what people think that they want when no one really even knows what they want right now." — Shensi Ding

Takeaway for operators: In emerging categories, companies that bring opinionated best practices to the table will win deals over companies waiting to respond to a defined spec.

6. Overlooked Insights

Insight 1: Identity Integration Is the Hidden Requirement for Agentic Governance

Buried in the Agent Handler discussion is a critical enterprise requirement that goes beyond permissions: tight coupling with identity providers so that employee offboarding instantly revokes all AI agent access. This is a quiet but significant enterprise procurement criterion.

Customers want Agent Handler to "tightly couple with our identity providers," Gil said, so that when an employee leaves, "they immediately get access revoked."

This suggests that any company building agentic tooling for enterprise will need to treat identity provider integration (Okta, Azure AD, etc.) as a first-class feature — not an afterthought — to close security-conscious buyers.

Insight 2: AI Buyers Are Skipping POCs Entirely

The traditional enterprise sales motion of proof-of-concept before commitment is being bypassed by AI-native buyers who move so fast they pre-validate before the first sales conversation.

Gil recalled customers responding to a proposed POC with "we already had our agent build into it. We know it works."

This has major implications for enterprise software go-to-market: the evaluation window is collapsing, which means product-led discovery (documentation, sandbox environments, community) may now matter more than traditional demo-driven sales cycles with AI-native accounts.