Investing in the SaaSpocalypse with Heller House's Marcelo Lima

Summary

• Marcelo Lima’s core call: the SaaS-pocalypse is a perception panic, and best-of-breed software is now priced for perpetual decline. Using 1/(R−G) with a 10% discount rate, 10x earnings implies zero growth forever — “a lot of these companies are there… or even priced for perpetual decline on a forward basis.” His names: Salesforce, ServiceNow, Atlassian, and Adobe (“extremely cheap”), with ServiceNow the only one he says isn’t priced for perpetual decline.
• The mechanism behind the call: Claude Code accelerates incumbents’ roadmaps rather than minting new competitors. AI will “not turn a poodle into a lion” — Volkswagen’s Project CARIAD lost billions trying to build software, PepsiCo’s competency is “selling snacks” — while Salesforce and ServiceNow deploy “armies of engineers” who just got dramatically more productive. The real enterprise barriers were never code cheapness but SOC 2 audits, governance, trust, and the fact that enterprise software “doesn’t get bought… it gets sold.”
• The moat that survives vibe coding is liability management, not features. A CTO ripping out SAP risks mission-critical downtime; Marcelo cites Jaguar-style attacks and says one or two hours of downtime can mean tens of millions in lost profit. You pay incumbents for SLAs and “one throat to choke.” The speakers’ AI-built tools, including Lima’s roughly 7,000-line internal project, still surface bugs despite high-effort prompts and repeated code-review requests — “it is not enterprise-quality software… no way.”
• Salesforce’s new headless/MCP release entrenches rather than commoditizes the platform, in Lima’s view. Stewart Butterfield told him in June 2019 “I never open Salesforce” — yet Salesforce mattered because 15–20 years of customer history, call logs, and permissioning sit underneath, data “Claude cannot see.” Counterintuitively, Lima’s own AI tooling “has now entrenched Excel because it depends on Excel existing.”
• Andrew Walker’s pushback stands unresolved: today’s AI tools are “the worst AI tools that we will ever use,” and Chegg also thought it was an AI beneficiary. Salesforce just announced “literally the largest ASR in history, $25 billion.” monday.com’s 20-F says it bought back 15% of its stock through the end of March; Walker inferred that, accounting for blackout periods, this was roughly a six-week window. Insiders started buying at 250 and 225 with stocks then at 180 and 160, and Lehman, First Republic, and Bed Bath & Beyond all had insider buying before zero. A VC also told Walker startups now delay their first Salesforce buy from Series B to “late Series C, early Series D.”
• The power law resolves the debate: a long tail of point solutions dies while entrenched platforms compound. Lima invokes Bessembinder’s finding that 4% of stocks created all net US equity wealth over 90 years; Walker adds a live tell — Domo’s fresh 10-K disclosed a first-ever CFO change-of-control agreement amid a strategic process, hours before recording.
• At the winners, engineer leverage means acceleration, not RIFs — but stock-comp math is a live risk. Lima argues you keep the 1,000 newly-10X engineers and “give them more tokens… crush our competition”; ServiceNow has pledged growth without headcount growth, and Marc Andreessen says engineers are working way more. Still, Atlassian spends over 100% of free cash flow on SBC buybacks — “completely absurd” — and Lima concedes “I don’t know” how underwater-comp turmoil resolves.
• Anthropic is positioning as the infrastructure layer, not the app killer. At its late-February enterprise event it said “we’re not here to destroy SaaS… [but to] become the infrastructure layer,” AWS-style; the precedent is MongoDB, whose stock crashed on Amazon’s DocumentDB and “then fully recovered and it’s up a lot.”

Deep dive

1. The January liquidation looked like sellers who never knew what they owned

• Lima’s vantage: he’s followed these companies closely since roughly 2018, was a software engineer in a previous career, and has been running Claude Code heavily while sending queries to all four leading models in parallel “since Gemini 3 Pro came out.” The incumbents were already shipping AI — ServiceNow’s Pro Plus AI SKU “went GA… might have been late 2023, don’t quote me on that” — so January’s shock read to him as a perception problem, not new information.
• The tape he’s fading: SaaS gapped down as semis gapped up in “alligator jaws,” with the Jefferies desk relaying “get me out at any price” and 10-to-1 sell-to-buy volume in software. His reaction: “did people not know what they owned?… Are they not aware that these companies are putting AI into their products already?”
• A self-aware caveat kept as said: “to my detriment, I’m so deep in the weeds that I don’t have the theory of mind to understand what other people are thinking.”

2. Claude Code arms the incumbents; it doesn’t turn a poodle into a lion

• The core logic chain: Claude Code makes developers far more productive, and developers work inside software companies. So AI “was always a tool that would accelerate the roadmap of existing software companies and not turn a poodle into a lion” — non-software companies don’t acquire software DNA “just because Claude Code came out.”
• His chosen counterexamples, as told: Volkswagen’s Project CARIAD — “a complete disaster, they lost billions of dollars” — and PepsiCo, whose core competency is “selling snacks.” His hedged concession: “you’re probably not going to see a PepsiCo vibe its own CRM, but you probably will see small companies vibe code their own CRM” for simple functionality.

3. Walker’s worst-case vs. Lima’s math: perpetual decline is already in the price

• Walker’s overarching worry: “the AI tools that we are using right now are the worst AI tools that we will ever use.” His topical exhibit is Anthropic’s Mythos — allegedly so good at zero-day exploits that the White House summoned bank CEOs — which he says instantly changed every cybersecurity company’s terminal value. Extend the logic: couldn’t Anthropic ship a Salesforce clone “80% as good… a heck of a lot cheaper”?
• Lima’s valuation frame: plug zero growth into 1/(R−G) at a 10% discount rate and you get 10x earnings — “a lot of these companies are there and priced for zero growth or even priced for perpetual decline on a forward basis.” He grants the multiple compression is “justified as a way to discount a lot of these worries.”
• The barrier was never code cheapness — cheap contractors in India, the Philippines, and Eastern Europe always existed, and a free open-source Photoshop (“I think it’s called PTGui”) has existed for decades while people pay for convenience. Enterprise barriers are trust, governance, and SOC 2 — “it would not pass an audit if you vibe coded the whole thing” — plus sales motions and Accenture-style integrators, because most enterprise software “gets sold.”

4. The Series B-to-Series D delay, and why it’s the incumbent’s prize to lose

• Walker’s anecdote from a VC: startups used to call Salesforce the day the Series B closed; now it’s “late Series C, early Series D.” His worry — small-company behavior is where the puck goes: extrapolate and it becomes “when we go public, or not at all.”
• Lima’s first question back: are they vibe-coding their own CRM, or “allocating budget to Anthropic instead” to ship an MVP faster? Walker concedes he didn’t press the detail.
• Lima’s framing: incumbents own the feedback loop — forward-deployed engineers inside their largest customers, enterprise go-to-market, daily deploys (Atlassian ships updates daily) — so “the onus is on them” to “provide so much value to the customer that it becomes irresponsible for that customer not to pay the monthly fee.”

5. Headless Salesforce: Walker sees a switchable back end, Lima sees entrenchment

• Salesforce 360 Headless — Benioff tweeted it that day — exposes Slack, Tableau, and MuleSoft as MCP “Lego bricks” so anyone can vibe-code a React dashboard on top. Walker’s worry: if Claude becomes the OS and owns the front end, “the back-end thing can be switched off pretty quickly” — or Anthropic just decides to take out a $100 billion company.
• Lima’s rebuttal is an anecdote from Slack’s June 2019 investor day, pre-acquisition: Stewart Butterfield told him “I never open Salesforce” — Slack updated it headlessly even then, and Salesforce still mattered because of 15–20 years of customer history, call logs, contracts, and governance “that Claude cannot see.” Same answer to Satya Nadella’s December 2024 “SaaS is CRUD” jab.
• The Openaw/OpenClaw sidebar: driving the agent tool with a free local Gemma 4 31B failed because “it hallucinates a lot”; on Sonnet he spent $2 “within two hellos” — every “hi” sends a million tokens of context, and he’s heard of people spending $200 a day. Lima says Salesforce open-sourced these context and guardrail files, which he thinks are called Agent Script — “just a bunch of text files.”
• His prediction, verbatim in spirit: “very soon you’re going to see both Windows and Mac… ship the operating system… with Openaw built in” and a local model, everything agentic and talking to MCP servers.

6. Chegg haunts the buyback signal — and the power law sorts survivors from Domo

• Walker’s pattern-match: Chegg came out in November 2023 saying it was an “AI beneficiary” and would use an ASR to buy back stock. Now Salesforce announces “literally the largest ASR in history, $25 billion.” monday.com’s 20-F says it bought back 15% of its stock through the end of March; Walker inferred, based on blackout periods, that this was roughly six weeks. Insiders who bought at 250 and 225 watched stocks fall to 180 and 160. “Do these guys really have a read on value, or is the landscape shifting so quickly underneath their feet?”
• The insider-buying graveyard, with Lima’s hedges intact: “I think” both Lehman and Bear Stearns had insider buying “literally the week before they went to zero,” “if I remember correctly”; First Republic had insider buying right before it went down as well. Bed Bath & Beyond was buying stock at $45 fifteen months before the death spiral. Lima’s conclusion: “every situation has to be evaluated differently.”
• Lima’s model of the outcome: Pareto distributions everywhere, capped by Bessembinder’s 4%-of-stocks finding. Point solutions may “not survive at all”; broad, mission-critical platforms with awake management — he credits Benioff’s “beginner’s mind” mantra — win. Asked for a change-of-heart name: “fortunately nothing that I own,” and he avoids decliners having “been burned so many times with value traps,” though Domo was “a SaaS loser five years ago.”
• Walker’s real-time governance dark arts: that morning Domo’s 10-K disclosed a first-ever CFO change-of-control agreement amid a strategic process; the CFO had been in the role since 2024 and at Domo since 2015. Domo had also agreed to let directors and employees settle bonuses in stock rather than cash. Walker compared the signal with Lionsgate granting its CEO a first stock-price-based RSU/PSU after 25 years.

7. Case-by-case evidence: Lightroom entrenches, Bloomberg churns, customers flip in 90 days

• Lima’s bull specimen: Adobe Lightroom’s AI on his 100-megapixel raw files — one button detects photo-bombing tourists and another makes them disappear, “not table stakes a year ago” — has made his back catalog more valuable. His bear specimen, counterintuitively: he left Bloomberg after endless data errors (a group chat with 19 others, about 20 people including Bloomberg employees, and two-to-three-day fixes) because a Claude script now updates his spreadsheets “flawless” — though he admits he’s atypical, not using Bloomberg chat.
• Lima’s counter-anecdote, worth keeping: customers of smaller software firms told him in October “we’re increasing our usage of this product because of AI,” then three months later “we stopped using it completely. The AI tools evolved and now we don’t even need them.” His takeaway: “if the customers can change that much in 3 months, I got to be really careful here.”
• On guidance, recorded April 17 ahead of earnings: Lima notes the barber-and-haircut problem with CEOs touting AI tailwinds and expects continued optimism because channel checks say enterprises go to existing vendors first — “can you agentify this thing for me?” — a refrain he hears from SAP and ServiceNow alike.

8. Stock-comp turmoil vs. the case for keeping all thousand engineers

• Walker’s mechanism: monday.com from 300 to 60 leaves options never vesting; making engineers whole turns his hypothetical roughly 2% annual dilution into 10%, and asking for “80% cuts in your stock comp” is a recipe for turmoil. Lima’s honest non-answer: “I don’t know, Andrew” — the 2022 analog (Meta down perhaps 70%, as he recalls, with comp repriced) didn’t obviously explode dilution, though “maybe it was painful and we’re seeing that right now”: Atlassian’s SBC buybacks exceed 100% of free cash flow, “completely absurd.” Under a new CFO, Lima says he is told Mike Cannon-Brookes now understands the need for a glide path to reduce stock comp as a percentage of revenue. Salesforce’s SBC is under a quarter of FCF; at ServiceNow only about 57% of FCF “is actually free.”
• Walker’s headcount math: if 1,000 engineers become 10X, maybe you need 100 or 300. Lima argues the opposite side, Munger-style: “let me keep them and give them more tokens so that we can accelerate our roadmap… crush our competition because we can run faster than anybody else” — and he thinks that’s what winners do, with net job creation in aggregate. ServiceNow has said it will grow without headcount growth.
• Supporting color from Marc Andreessen’s Harry Stebbings interview: engineers are working “way more,” not less. Both speakers confess the Limitless-drug effect — Lima woke up excited to work and found it was 2:00 a.m.

9. The real moat is liability — and Anthropic wants to be AWS, not the app

• The “one throat to choke” argument: a PepsiCo CTO ripping out SAP risks mission-critical downtime. Lima says the invoice buys “not only the product… liability management”: SLAs, guaranteed nines, and a support operation that can send roughly 100 engineers and perhaps restore service in 30 minutes to two hours, versus weeks for a small team using a vibe-coded replacement. Walker adds that airlines and hospitals can face especially severe consequences from downtime.
• Vibe code’s ceiling from their shared experience: Lima’s roughly 7,000-line internal tool still surfaces new bugs daily. Andrew says even after asking Opus 4.6, with high effort, for mistake-free work and running repeated full code reviews, his own tools continued to find bugs, including a bogus 52-week-low alert. “As amazing as I think it is, it is not enterprise-quality software… no way.”
• The closing frame: at its late-February enterprise event Anthropic said “we’re not here to destroy SaaS. We’re here to partner with… software companies and become the infrastructure layer” — the AWS analogy, providing intelligence as AWS provides primitives such as storage and compute. Precedent: Amazon’s DocumentDB attack on MongoDB, whose stock “crashed and then fully recovered and it’s up a lot.” Hedge preserved: “we’ll see what the industry structure looks like.”