Newsletter · · Ashutosh Agarwal

Salesforce Downgraded as Agentforce Adoption Stalls - Is SaaS Broken? - Week of July 10, 2026

Software and internet newsletter for the week of July 4 to 10, 2026. KeyBank cut Salesforce to Hold on stalling Agentforce adoption and Benioff hit back on live TV, as the tape split software into AI infrastructure winners like Datadog and seat-priced application losers like Salesforce and Adobe.

Is SaaS Broken?

Week of July 10, 2026: Salesforce Downgraded as Agentforce Adoption Stalls

Last week the story was customers pushing back on usage-based pricing. This week the pressure moved from the pricing model to the product itself: for the first time, the slowdown in Salesforce's flagship AI product showed up as a hard sell-side downgrade, not just chatter. Underneath it, the same question keeps getting louder: if one AI agent can do the work of ten people, who keeps paying for the other nine seats?

TL;DR

  • Salesforce (CRM) took its first real bruise of the AI era: KeyBank cut it from Buy to Hold specifically because Agentforce adoption is slowing. CEO Marc Benioff went on the record calling the downgrade "wrong." The stock is now the worst-performing name in the Dow this year, down about 40%.
  • The market is splitting software into winners and losers by role: the "infrastructure" names that help run AI systems (Datadog is up 88% year to date) are being bid up as clear winners, while the "application" names people log into all day (Salesforce, Adobe) are where the bills are coming due.
  • The seat-erosion math got blunt: leaked OpenAI financials show it still spends $1.60 to earn a dollar, a research firm now pegs a fifth of all software subscription spending ($234 billion) as "in play" by 2030, and Notion quietly killed one of its own apps because AI agents had made it unnecessary.

What's New

Ranked by what actually matters for positioning a book this week.

1. Salesforce's AI Story Finally Cost It a Rating, and Benioff Fought Back on Live TV

This is the week's most actionable single event. On CNBC's Squawk on the Street (July 9), Jim Cramer walked through KeyBank analyst Jackson Ader's decision to downgrade Salesforce from Buy to Hold, "going from difficult to find evidence of future upside." The reason was specific: Ader "sees slowing adoption in Agentforce, which is really... that was going to be the future." Cramer's own gloss was harsher: "Agentforce, proof of concept, just not happening."

Why this moves numbers: Agentforce is the product Salesforce has told investors will re-accelerate growth. If independent field checks show adoption stalling, the whole re-rating case wobbles. Cramer also spelled out the budget mechanism that should worry any per-seat software holder: the people who set corporate budgets "say, look, let's see if we can't not spend as much money on Salesforce, which they think is expensive... let's try to implement Agentforce, but let's not go too fast because we may not get it right." That is exactly how a growth story turns into a spending target.

To his credit, Benioff didn't take it quietly. Cramer relayed that the CEO pushed back on the record, pointing to "100 references about how Agentforce is doing" and flatly calling the analyst's take "wrong." Notably, even the bearish analyst "doesn't disagree" with Salesforce's giant buyback or its "very big target out in 2030." So the debate isn't whether Salesforce survives, it's whether the AI product delivers on schedule. The stock has already voted: "there's no Dow stock that's been worse so far this year," down roughly 40%.

2. Wall Street Is Now Sorting Software Into Two Piles, and Datadog Is in the Winner Pile

On Bloomberg Surveillance (July 8), a sell-side software analyst laid out the cleanest framework of the week: "We're continuing to see the bifurcation in software. So you have the two neighborhoods, the application software names" (his examples were "the Salesforce.coms and the Adobes") "and then... the infrastructure neighborhood."

The verdict is lopsided. Infrastructure software "continue[s] to get bid up... the most correct way to express the veritable AI winners in the software stack." His marquee example was Datadog, which he described in plain terms as "the equivalent of an MRI scan for your entire IT architectural topology": the more companies wire together tangles of AI agents that can "hallucinate," "spin up a wrong answer," or "outright fail," the more they need something watching the whole system. His point on the stock: "it's up 88% year to date." Application software, by contrast, was "looking fat and happy. And now... those bills are coming due" (he noted Microsoft is "down 25% year to date").

Why this moves numbers: this is the market pricing the "Is SaaS broken?" question by category. The tools you sit inside all day (a Salesforce screen, a Photoshop window) are seen as exposed to seat erosion; the plumbing that measures AI systems is seen as a picks-and-shovels bet that grows with AI adoption. For a book, that argues for owning the observers of AI complexity and being cautious on the seat-priced apps. A second guest, Sarah Hunt of Alpine Saxon Woods, put the core worry simply: "if you're growing your... user base and that starts to shrink, that's a problem for the cash flow."

3. Leaked OpenAI Numbers Make the Seat-Erosion Threat Concrete: a Fifth of Software Spend Is in Play

The Business of Tech (July 7) episode connected the dots better than anyone this week. Host Dave Sobel cited "OpenAI's own audited financials, leaked and verified by the Financial Times," showing the company "spent $1.60 for every dollar it earned. And that was the good news. The year before, it was $2.37." In plain English: the company selling the AI engine still loses money on every dollar of software revenue it books, so it can't afford to wait around, and is racing toward services instead.

He paired that with a striking research-firm estimate: Gartner projects that "agentic AI could affect $234 billion of SaaS spending by 2030. Roughly 20% of everything businesses spend on software subscriptions." The mechanism is the one this newsletter keeps circling: "the seat, the license, the per-user-per-month line was built on the person sitting in the application. The agent doesn't buy a seat."

The most vivid data point was a company killing its own product on purpose. Sobel noted Notion shut down Notion Mail, its email app, because "more than half of the people using Notion Mail were managing their email without ever opening the inbox view. The agent handled it." He also flagged where the big players are reallocating: Microsoft is standing up a $2.5 billion "Microsoft Frontier" unit with 6,000 embedded consultants, Amazon committed $1 billion to a similar model, and OpenAI launched a $150 million partner program aiming for "300,000 certified consultants by the end of the year." The giants that invented per-seat software are sprinting toward people-and-outcomes revenue.

4. The Pricing Pivot Has a Named Operator Now, and a Clear Shape: Hybrid, Not Pure Usage

Last week's cautionary tale was Help Scout trying to move off per-seat pricing and finding customers preferred the predictability of seats. This week added the other half of the picture. On The GTMnow Podcast (July 9), Notion's revenue-operations lead Brian Le said the company's "big focus is around shifting from our traditional SaaS seat-base, license-base model to unlocking this new aha moment on how do we actually sell efficiency? How do we actually sell work?" He added that "a lot of customers and other players in that space [are] moving to usage-based pricing... this consumption model because we're in the era of efficiencies."

Stripe's AI chief gave the industry-wide version on The MAD Podcast (July 9). His explanation of why the old model breaks is the tightest statement of this whole thesis: classic SaaS had "beautiful and simple to monetize economics... marginal costs are near zero. And that's why SaaS margins are really good. And that's why... seat-based licenses work really well." AI "breaks that model because... every prompt and every API call and every task has a real marginal cost all of a sudden... the inference isn't free." His field read: "I see very few scaling or scaled AI companies that are still exclusively subscriptions or seat-based." But, and this reconciles it with last week, the winning shape isn't pure metering; it's a hybrid, a familiar subscription with usage-based billing kicking in above a threshold (he cited Lovable and 11 Labs both migrating this way). Why it moves numbers: a hybrid model keeps some revenue predictable (good for the multiple) while tying the rest to consumption (good for margins if inference costs behave). It's the most realistic middle path between "seats forever" and "usage breaks everything."

5. Adobe's Problem Isn't Just AI, It's an Empty C-Suite and a Voluntary Hit to Subscription Revenue

On InvestTalk (July 8), host Luke Guerrero of KPP Financial ran through an Adobe picture that is grim on governance. "Their CEO is stepping down after 18 years... They have no permanent CEO. They have no permanent CFO who stepped down on the 15th." Worse for the thesis, he said Adobe "did this pivot because they're really concerned about their AI user growth... towards freemium. So they're explicitly accepting pressure on subscription revenue": a company choosing to trade paid seats for free users to defend against AI-native rivals.

The market reaction has been brutal even against genuinely good results: the stock is "down 41% over the past 52 weeks... down 36% year to date... in spite of having a record quarter. 13% year-over-year growth... AI-first ARR triple year over year... beat on earnings by about 2.4%." That leaves it "at eight times forward-looking price earnings... 84% return on equity." Guerrero's honest verdict captures the melting-ice-cube worry: it may be "too cheap to sell," but with "AI-native tools eroding Creative Cloud" and no leadership in place, there's no near-term catalyst to buy either.

The Debate: Is Per-Seat SaaS Structurally Broken, or Just Re-Rating to Consumption?

The bear case (it's breaking). The unit that SaaS was built on, a human paying per month to sit inside an app, is being detached from the work. Ryan Morris, a channel consultant, put the mechanic bluntly on Business of Tech (July 9): "if you use an AI agentic model and it can do the work of all of these people, they just went from 10 authorized users who need access to a system to one. And now the other nine are doing something else. But why would they continue to pay you for people who are no longer needed?" And once buyers pay by consumption, they behave the way they behave with any utility bill: "I will cost-optimize against the consumption model 100% of the time." Add the macro backdrop from Prof G Markets (July 6): OpenAI made "$13 billion in revenue... but they spent $34 billion," a "$21 billion operating loss," with free Chinese models jumping "from 30% of AI traffic to 60% in six months," and you get an industry where even the engine-maker loses money and price pressure only builds. Datadog's 88% run versus Salesforce's 40% collapse is the market already pricing the split.

The bull case (it's re-rating, not dying). The steel-man came from a serious investor. On Private Equity Spotlight (July 8), Dame Anne Glover of Amadeus Capital Partners called the roughly "$300 billion... wiped out from global markets" an "overreaction." Her nuance is the crux of the whole debate: SaaS companies won't die because AI replaces their engineers, "they should do that anyway, because that's just becoming more efficient." They'll die only if "they're not understanding the customer need and someone inside the organization can do it faster and better." In other words, this is ordinary innovation disruption: the well-run, customer-obsessed incumbents adapt; the lazy ones become "dinosaurs." She's blunt that "it does not replace humans. It just makes humans more productive." The a16z team backs the stickiness half of this: on The a16z Show (July 7), Steven Sinofsky argued "almost everything interesting in an enterprise is an exception," and the deep business logic codified in tools like SAP and Salesforce means "you can't just... vibe code your way into enterprise software." Ryan Morris, the same bear from above, even conceded "we're so early in the process that we have time to adapt," citing BCG's rule that only 10% of the value is the technology, 20% the data, and 70% the humans and processes around it.

Where the swing sits. Both sides now agree on the destination, some blend of seats plus consumption plus outcome/services revenue, and disagree only on the speed and on who adapts. The single most important unresolved fact is still the one this newsletter has flagged for five straight weeks: no in-scope SaaS company has yet disclosed an explicit gross margin on its AI features, and none has printed a fresh net-revenue-retention number (the key measure of whether existing customers spend more or less over time). Until those land, the bull and bear cases are arguing over a number neither can see.

Stocks in Play

Note: this week's podcasts gave direct, substantive airtime to CRM, ADBE, and DDOG. TEAM, HUBS, ASAN, and MNDY were essentially dark, so their read is carried by the seat-erosion theme, flagged honestly below.

Adobe (ADBE)

  • Bull: Record quarter, 13% revenue growth, AI-first ARR tripling year over year, an earnings beat, 8x forward earnings, 84% return on equity (InvestTalk, July 8). Arguably too cheap to sell after a 41% one-year drop.
  • Bear: No permanent CEO or CFO, a deliberate pivot to freemium that "explicitly accept[s] pressure on subscription revenue," and AI-native design tools nibbling at Creative Cloud. Down 36% year to date despite good numbers: the market doesn't believe the earnings can hold.
  • Next catalyst: A permanent CEO/CFO appointment, and whether the delayed price actions land, plus any hard number on how freemium dents paid subscriptions.

Salesforce (CRM)

  • Bull: Benioff insists Agentforce traction is real ("100 references"), the buyback is huge, and even the bear analyst backs the 2030 target (Squawk on the Street, July 9). The stock is cheap after a ~40% drop.
  • Bear: KeyBank cut it to Hold on "slowing adoption in Agentforce"; budget owners are actively looking to spend less on Salesforce; and a16z argues its "Headless 360" launch was "a marketing announcement... nothing actually changed" (The a16z Show, July 7).
  • Next catalyst: Any concrete Agentforce adoption or consumption metric on the next earnings call: the single number that would settle the Cramer-vs-Benioff fight.

Datadog (DDOG)

  • Bull: Explicitly named the "veritable AI winner" among infrastructure software, the "MRI scan" for increasingly fragile AI-agent systems, up 88% year to date (Bloomberg Surveillance, July 8). Its consumption-based model is a feature here, not a bug: more AI usage means more to monitor.
  • Bear: An 88% run prices in a lot of good news. The same consumption model that helps on the way up amplifies any slowdown in customer AI activity, and the InvestTalk read that the token-usage index is rolling over (see Read-throughs) is a warning that AI activity growth can stall.
  • Next catalyst: Usage/consumption trends and any commentary on LLM-observability revenue on the next print.

Atlassian (TEAM), quiet this week

  • Bull: Last week's story still stands: its "teamwork graph" is a context-and-data moat that agents need, hard to replicate.
  • Bear: Pure read-through from the seat-erosion theme: Atlassian is a per-seat, teamwork-tool vendor, exactly the profile Ryan Morris and Gartner describe as exposed if agents collapse seat counts.
  • Next catalyst: A Rovo adoption or consumption metric, and whether opening its graph to rival agents dilutes its own seat capture.

HubSpot (HUBS), quiet this week

Only an ex-executive's general go-to-market discussion this week, nothing on Breeze AI.

  • Bull: SMB breadth and an integrated suite; Breeze AI could lift revenue per customer if it lands.
  • Bear: SMBs are the customers most able to "hack around" with cheap agentic tools, and HubSpot's per-seat core sits squarely in the erosion zone.
  • Next catalyst: A first Breeze monetization or attach-rate figure; any net-revenue-retention update.

Asana (ASAN), dark this week

  • Bull: Management has already conceded the low end is exposed and is moving upmarket to larger enterprises, arguably the right defensive move.
  • Bear: Work-management seats are a textbook "one agent replaces five seats" target; silence in the podcasts isn't reassurance, just absence of new information.
  • Next catalyst: Any AI Studio usage data or enterprise-cohort retention number.

Monday.com (MNDY), dark for a second straight week

  • Bull: AI-credit consumption pricing is, in theory, the model best positioned for the consumption era: it could turn AI usage into upside rather than seat loss.
  • Bear: We still have zero fresh data. After being the only directly named in-scope stock two weeks ago, MNDY has gone completely silent: no AI-credit consumption metric to judge whether the model is working.
  • Next catalyst: A first disclosed AI-credit consumption or attach number.

Read-throughs

Seat-heavy SaaS (TEAM, HUBS, ASAN, MNDY). These four got little or no direct airtime, but they're the purest expression of the week's central risk. Ryan Morris's "10 users to 1" example and Gartner's "$234 billion / 20% of software spend in play" estimate are aimed straight at per-seat collaboration and work-management tools. The offsetting hope is the hybrid pricing path (Notion, and the Stripe framework): if these vendors can convert seat pricing into a subscription-plus-consumption blend before agents thin out their seats, they re-rate rather than break. Watch which ones announce a pricing change first.

Model and inference vendors (OpenAI, Anthropic, Bedrock, Azure, DeepMind). The economics stayed ugly and got more concrete: OpenAI at $1.60 spent per dollar earned and a $21 billion loss, Anthropic estimated around an $11 billion loss, and free Chinese models taking share fast (Prof G Markets, July 6). One hopeful thread continued from last week: Everyday AI (July 6) reported "OpenAI has found a way to cut inference costs in half. And if that's true, that is a monumental deal." For SaaS, cheaper inference is the difference between AI features being a margin drag and a margin-neutral upgrade, but as noted last week, vendors may simply keep the savings. The same episode flagged Microsoft banning internal use of Anthropic's newest models over data-retention terms, a reminder that enterprise trust, not just price, gates model adoption.

Multiple de-rating risk. The InvestTalk episode pointed to a concrete measuring stick: the Silicon Data LLM Token Expenditure Index, a real-world gauge of what users pay for AI, is "down almost 20% from its May high after nearly doubling since its inception in December." If that keeps falling, it signals eroding pricing power across the whole AI stack, which pressures both the model vendors and the consumption-priced software (like Datadog) that rides on AI activity. Anne Glover's warning stands as the sober backdrop: there "will be an AI crash," she just "[doesn't] know when."

What Changed vs. Last Week

New: Salesforce moved from footnote to headline. Last week CRM appeared only as a read-through. This week it's the lead story with a named, dated sell-side downgrade (KeyBank, Buy to Hold) tied specifically to Agentforce adoption, plus a public Benioff rebuttal. That's a genuine escalation.

New: Datadog got its first direct coverage in weeks, and it was bullish. DDOG had zero coverage last week; this week it was singled out as the archetypal AI winner (up 88% year to date) inside a clean "application vs. infrastructure" framework. That framework itself is new and useful for sorting the seven names.

Corroborated: the OpenAI cost story got two fresh, harder data points. Last week: OpenAI reportedly cut inference costs "by more than half" but plans to keep it as margin. This week: leaked audited financials ($1.60 spent per dollar, improving from $2.37) and the $13B-revenue / $34B-spend / $21B-loss figures. The margin-floor thesis is now better documented, not resolved.

Evolved: the pricing debate found its synthesis, hybrid. Last week's Help Scout story (customers rejecting pure usage pricing, reverting to hybrid) now dovetails with a named operator (Notion) actively shifting toward consumption and Stripe's read that almost no scaled AI company is seat-only, with hybrid (subscription + metered overage) as the emerging norm. The two weeks fit together rather than contradicting.

Still missing (five weeks running): the two numbers that would end the debate. No explicit AI-feature gross margin percentage from any of the seven in-scope names, and no fresh net-revenue-retention print from any of them. Anthropic's 38% to 70% inference-margin claim from two weeks ago is still uncorroborated.

Quiet: four of seven tickers went dark. TEAM (a direct name last week via its teamwork graph), HUBS, ASAN, and MNDY had no substantive coverage this week. MNDY has now gone silent for two straight weeks after being the only directly named in-scope stock a fortnight ago.