AI Agents Are Rewriting the SaaS Stack

Enterprise AI spending is growing far faster than most public commentary suggests.

Menlo Ventures' estimates imply a jump from roughly $33B to nearly $177B in enterprise AI spend within a year, signaling that adoption is moving from experimentation to large-scale operational deployment. This matters because it shows AI is becoming a major budget line item, not just a pilot project. The scale of this increase also suggests that enterprise demand for AI infrastructure, APIs, and applications is compounding quickly.

The core shift is from buying seats to buying tokens, and agents are the main reason.

Instead of paying for a fixed number of users, companies are increasingly paying for usage-based token consumption tied to workloads, especially agentic systems that can operate autonomously and in parallel. This is significant because it makes spend scale with task volume rather than headcount, which can drive much higher and less predictable costs. Technologies like agent workflows, token-based pricing, and automated coding assistants are central to this change.

Enterprises are favoring frontier models over cheaper alternatives more than before.

The report indicates frontier model spend share rose from 35% to 63%, suggesting that buyers are prioritizing capability over low-cost options when the use case is important enough. This is notable because it implies that model quality, reliability, and performance still dominate procurement decisions in high-value enterprise settings. It also reinforces the competitive importance of leading closed models from providers like Anthropic and OpenAI.

Anthropic has emerged as the leading enterprise API provider, while OpenAI has rebounded and Google has slipped.

The episode emphasizes that Anthropic now appears to be the dominant force in enterprise API spend, with OpenAI recovering some share and Google losing a meaningful portion. This is important because it reflects how quickly the enterprise market can re-rank suppliers based on product quality and developer preference. It also highlights the strategic value of frontier model performance in areas like agentic coding and enterprise automation.

A large share of AI spending is net new, not just repurposed from SaaS budgets.

The data suggests AI spending is expanding total enterprise budgets rather than simply replacing existing software line items. That weakens the simplistic 'AI kills SaaS' narrative, at least in the near term, because many organizations are layering AI on top of current tooling instead of substituting it. This has implications for software vendors, investors, and CIOs trying to understand whether AI is a cost center, a replacement, or an expansionary technology.

Token share may be a better indicator of future market direction than spend share alone.

Open-source models may represent a much larger share of token usage than revenue, meaning they are heavily used even if they are not monetized at the same level as frontier models. That distinction matters because token volume can reveal where actual workload is flowing, not just where money is spent today. It also suggests that the economics of AI adoption may evolve differently from the current revenue distribution among model providers.

The data-center backlash is likely to soften into negotiation rather than outright rejection.

Jason argues that once elections are over, politicians tend to shift from rhetoric to governance, especially when jobs, tax revenue, and infrastructure investment are at stake. This matters because the future of AI infrastructure will probably be decided through regulation, incentives, and local compromises rather than bans. Concepts like electricity supply, water usage, and community benefits become the real bargaining points.

Jason sees Trump as having been unusually strong on AI infrastructure and energy policy.

He frames Trump’s stance as broadly pro-growth, especially around backing data centers and energy expansion to support AI. The significance is that AI policy is becoming a partisan signal, but the practical issue is whether leaders support the compute and power buildout needed for scale. This includes the broader ecosystem of data centers, grid capacity, and industrial energy policy.

Market fears around AI and data centers are presented as an overreaction.

Jason says the post-election dip in tech stocks is a buying opportunity rather than a sign that the AI story is breaking down. He specifically points to Nvidia, Amazon, Uber, and Tesla as names he would buy on weakness. The takeaway is that he still views AI infrastructure and adjacent platforms as structurally important, not temporary hype.

AI content licensing and training rights remain unresolved, but the industry is moving toward paid deals.

The episode distinguishes between stolen content, terms-of-service violations, and the broader legal question of whether training models on published material is allowed. This is important because it suggests the business model for AI may shift toward licensing and revenue-sharing with publishers and rights holders. Technologies and concepts involved include model training, output attribution, fair use, and content licensing agreements.

Apple’s AI spending cuts may be a sign that compute costs are becoming operationally real.

Jason interprets Apple’s reduced AI budgets as evidence that token usage and compute consumption are expensive enough to force tighter discipline. That could push companies toward usage caps, more selective product rollout, and possibly local or hybrid model deployment. This is significant because it links AI strategy directly to unit economics, not just product ambition.

Democrats and tech leaders may be headed back toward a working relationship.

The hosts suggest that post-election politics will likely move from confrontation to compromise, especially around AI and data-center development. Jason’s view is that Democrats should bring tech CEOs back into the conversation instead of treating them as permanent adversaries. This matters because policy outcomes in AI will likely depend on collaboration among government, enterprise, and infrastructure providers.

Salesforce evaluates acquisitions through a clear three-part lens: strategic fit, acceleration of the business, and reasonable value creation.

Somorjai frames M&A as more than buying technology; the deal must meaningfully improve Salesforce’s roadmap or market position. This framework helps explain why the company has been selective despite its large acquisition history, including deals like Slack and MuleSoft. It also shows how enterprise acquisitions are judged on business impact, not just product novelty.

Enterprise software pricing is shifting toward consumption and outcomes rather than per-seat licenses.

As AI reduces the amount of human labor required, seat-based pricing becomes less aligned with customer value. Salesforce’s mention of consumption-based models, such as MuleSoft’s, highlights how pricing is adapting to usage and results. This matters because it affects how AI products are monetized and how enterprise vendors prove ROI.

Distribution is a major acquisition advantage: Salesforce can give startups immediate access to a massive sales and customer network.

One of the clearest reasons to sell to Salesforce is the ability to plug a product into 15,000 sales reps and a large installed base almost overnight. That kind of distribution can dramatically accelerate adoption, revenue, and credibility for a startup. For founders, this means an acquisition can be a growth strategy, not just an exit.

Slack has become even more strategically important in the AI era because it serves as the conversational interface and context layer.

Somorjai argues that Slack is not just a communication tool—it is becoming the user interface for humans and AI agents interacting with Salesforce products. This makes Slack central to how users will experience workflow automation and AI-driven enterprise software. The episode suggests that conversational interfaces may matter as much as backend capability in the next generation of software.

Salesforce Ventures is broadening from enterprise software into frontier AI, voice, video, and robotics.

The move reflects a belief that the next wave of enterprise value will come from AI-native systems and adjacent frontier technologies, not just traditional SaaS. The mention of a $1 billion AI fund underscores how seriously Salesforce is leaning into this shift. It also signals that corporate venture capital is being used to scout and shape emerging markets.

Diligence can kill a deal, especially when there are financial, contract, or security issues.

Somorjai notes that weak controls, poor customer data, and security vulnerabilities are major red flags, with serious security problems able to stop a deal outright. This highlights that M&A is not only about vision and strategy; execution risk and operational quality matter heavily. For founders, this is a reminder that clean operations can materially affect exit outcomes.