Saaspocalypse: why 10,000 new pieces of software a day is not an exaggeration

By Paul-Antoine TUAL, AI Transformation Leader, Croissance et Transitions, August 2026.

Framing. A number circulating online deserves the same treatment as a stock-market number: you verify it before you quote it, you do not judge it by how it looks. We were handed the claim that the world creates at least 10,000 new pieces of software a day, presented as a conservative estimate. It is, and even more conservative than its author thought. But the striking fact of 2026 is not that number: it is that the stock market drew the conclusion before we did. We had already tracked the AI maturity of the CAC 40 and the Nasdaq, then AI’s money through its value chain, then the signals of a speculative bubble. Here, a fourth market: software itself, grown so abundant that it wiped out more than $1 trillion in stock-market value from those who sold it by the seat. This article is not investment advice; several figures are press estimates or aggregator methodologies that diverge from one another, flagged as such.

Update, 7 August 2026. Two days after this piece was published, Airtable was sold at roughly 11% of its peak valuation, with its AI agent platform legally carved out of the deal before signing. The case is covered in the addendum, at the end of the article.

---

1. 12 January 2026: one launch, six weeks of vertigo

The signal was not new. Back in late 2024, Swedish fintech Klarna had already announced it had ditched Salesforce’s flagship CRM for a home-grown, AI-powered system [9]. An isolated case, at the time. What changes in 2026 is the scale and the speed.

On 12 January 2026, Anthropic launches Claude Cowork: an agentic, no-code version of Claude Code, in early access for Pro subscribers. The tool no longer targets developers alone. It promises any office worker the ability to automate whole tasks that used to be handed to software sold by the seat [1]. On 30 January, eleven industry-specific modules follow, for fields such as cybersecurity or legal work [1]. The market takes a few days to digest the news, then reacts.

In early February, Reuters puts the stock-market value wiped from software and services shares at close to $1 trillion, over the first six weeks of the year [2]. On 6 February, Forbes confirms a comparable order of magnitude: the iShares Expanded Tech-Software ETF (IGV) falls about 21% while the S&P 500 holds steady, and Salesforce loses 26% [3]. That is the day a trader, quoted by Forbes, names the episode the “Saaspocalypse” [3]. On 24 February, software stocks partly rebound after Anthropic announces new partnerships with several vendors [4]. The relief is short-lived: over the rest of the year, the erosion resumes.

The market did not take six months to understand. Six weeks was enough.

2. Seat compression: why classic SaaS is taking the hit

The mechanism has a name in the financial press: “seat compression”. An AI agent able to do the work of five people does not save the client company time. It makes it buy one licence instead of five.

Forbes cites three indicators that give the mechanism substance [3]. The average number of SaaS applications per company is said to have fallen from 112 to 106. 82% of companies surveyed say they are actively cutting the number of software vendors they use. And the share of vendors billing strictly per seat is said to have dropped from 21% to 15% in twelve months, as each one hunts for usage-based or outcome-based pricing to survive the shift. One market analysis goes further on valuation: the software sector’s forward earnings multiple, around 84 times at its 2020-2022 peak, is said to have fallen to about 22 times by March 2026 [8]. That figure deserves caution, it comes from a market newsletter rather than a statistical body, but its direction matches everything else.

It certainly matches Salesforce’s trajectory, the most-cited name in this story. On 6 January 2026, the stock is down 3.5% year to date, an unremarkable move [5]. On 6 February, after the Claude Cowork shock, the cumulative decline reaches 26% [3]. On 27 April, it stands at 31%, alongside Adobe, which shows the same figure [6]. On 21 July, the stock trades around $152, down 43% since January; Adobe is past 35% and Workday has reached 45.5% [7].

3. What 10,000 pieces of software a day actually means

Here is the text submitted to us for this article: the world would create at least 10,000 new pieces of software a day, said to be a conservative estimate. We checked it platform by platform rather than taking it at face value. The floor holds, with mid-2026 figures noticeably higher than those originally put forward.

Take Lovable, the best-known of the natural-language app generators. On 8 March 2026, its users create or update 500,000 projects in a single day, a peak; the average at the time sits closer to 200,000 a day. Three months later, in June, TechCrunch reports the company has passed one million new projects a week, alongside $500 million in annualised revenue [10]. That works out to an average of roughly 140,000 a day, a weekly cadence, not to be confused with March’s peak. The two figures do not contradict each other, they measure different things: a one-off high versus a cruising speed, on a platform founded in late 2023 that has not yet reached its third birthday [10][11].

Traditional development is not being left behind. Over the period covered by its Octoverse report (September 2024 to August 2025), GitHub logs 121 million new repositories, more than 230 a minute, carried by over 180 million developers, of whom 36 million signed up in that year alone [12]. A growing share of that flow is itself written by AI: according to a SemiAnalysis analysis picked up by several trade outlets, Claude Code alone was authoring around 4% of GitHub’s public commits in early February 2026, a share more recent tracking puts closer to 9 to 10% by mid-year, on a trajectory projected past 20% by the end of 2026 [13].

At the other end of the scale, the strictest definition of software, a validated, published product, yields smaller but still far from negligible volumes. Mobile app stores add roughly 1,249 new iOS apps a day on one 2026 estimate, and between 1,200 and 1,250 new Android apps, a figure that climbs past 2,400 depending on the counting method used [14][15]. Stripe, for its part, was onboarding roughly 1,000 new merchant businesses a day in 2025, with competing estimates running as high as over 2,000 [16]. Added together, app stores and Stripe cap out around 3,500 a day, below the 10,000 mark. That is precisely what makes the question interesting: the 10,000 floor is not cleared by finished software in the classic sense. It is cleared the moment you count generation, code repositories and web prototypes included.

The floor is not 10,000. It is far higher.

4. The honest counter-argument: creating is not keeping

It is worth resisting the pull of a single grand narrative. Two caveats matter here, and both hold up.

The first concerns the market itself. Not every SaaS company faces the same odds against AI. Forbes distinguishes vendors selling simple features, replaceable in a weekend of vibe coding, from vendors deeply embedded in complex business processes, where proprietary data and network effects remain a real moat [3]. Analyst Dan Ives cites Microsoft, Palantir and CrowdStrike as relatively spared. The rebound of 24 February, following Anthropic’s announcement of partnerships with several vendors, confirms that the market is not betting on wholesale extinction, but on a sorting process [4].

The second caveat concerns the quality of what gets created. We wrote about this in our article on the end of vibe coding: when a model has a choice between a safe and an unsafe way to write a piece of code, it picks the vulnerable version 45% of the time, a rate that has stayed stable across every model family tested between 2025 and 2026. Generating 330,000 repositories and 140,000 projects a day says nothing about their reliability, their upkeep, or their security. The Saaspocalypse is not the end of paid software. It is the end of its scarcity. And a scarcity that disappears on one side always makes another appear elsewhere: we had already documented that shift, from generation towards orchestration and verification, in our software publisher playbook.

Creating is no longer the problem. Keeping is.

5. Three pointers for a business leader, not a trader

An SME or mid-cap leader does not have to choose whether to buy or sell software stocks. But this market story offers three concrete pointers for their own information system.

Every line on the software bill now deserves a standing question: “what if we built this ourselves, in a week, with an AI agent?” That is not always the right answer, proprietary data and deep integration remain real moats, but the question now has to be asked for every line. The abundance of generated code also shifts value towards verification rather than production, exactly what we detailed in UltraCoding: a leader who invests in review, audit and a quality bar protects their company better than one who invests only in generation speed. How software gets bought matters just as much as the software itself: that is the whole point of our article on the sovereign business base rather than reselling tokens, an answer built before the word Saaspocalypse even existed.

---

Addendum, 7 August 2026: Airtable, the demonstration in three days

This article was published on 2 August 2026. On 4 August, Italy’s Bending Spoons signed the acquisition of Airtable. The Form 6-K filed with the SEC that same day puts numbers on the deal: $1.285 billion in enterprise value, entirely in cash, implying an equity value of roughly $2.25 billion once the net cash on the balance sheet is added back [20]. Closing is expected before the end of 2026, with a long-stop date of 4 February 2027 [20].

Airtable was worth $11.7 billion in December 2021, at the close of a $735 million Series F [23]. Its shares were still changing hands at around $4 billion on the secondary market in early 2026, after more than $1.4 billion raised in total [21].

Two readings of the same price have been circulating for three days, and both are accurate. Semafor goes with “roughly 10% of its peak valuation”: that is the enterprise value, 1.285 against 11.7, or 11% [22]. Other commentators go with nineteen cents on the dollar: that is the equity value, 2.25 against 11.7, or 19%. The gap between the two, roughly $965 million, is the cash Airtable holds and the buyer picks up along the way. So the 89% fall applies to the operating business alone, the 81% fall to the business plus the cash.

What makes this price relevant here is its nature. A share price is a revisable opinion, produced daily by anonymous sellers who have never opened the books. Here, two parties reviewed the accounts, hired lawyers and signed. The 2.7 times annual recurring revenue ($1.285 billion against roughly $480 million of ARR in June 2026, growing more than 20% year on year) is a negotiated price [20][23]. The median listed B2B software company trades at around 4.1 times ARR [23].

The clause the press release does not lead with

The Form 6-K contains one sentence that carries as much weight as the price: “Prior to entering into the Purchase Agreement, Seller and its affiliates implemented a reorganization pursuant to which […] assets and liabilities relating to the “Hyperagent” business line were transferred by the Company to Hyperagent Inc.” [20]. Before signing, Airtable moved its AI agent platform out of the perimeter being sold, into a separate Delaware corporation (see Figure 4).

The buyer takes the database, the 500,000 customer organisations, 80% of the Fortune 100, the $480 million of recurring revenue and the cash [20]. The founders keep the agents. Two assets, two prices, one contract, on the same day, between the same people. This is the argument of section 2 written up as contractual clauses: the per-seat SaaS body and the agentic promise were valued so differently that they had to be legally separated before the first one could be sold.

Airtable bills per seat, $20 per user per month on its Team plan, $45 on Business [24]. The exact profile described in section 2, word for word.

What the deal does not prove

Three caveats, better raised here than left for a reader to find.

The starting valuation was itself excessive. Airtable roughly tripled its recurring revenue between 2021 and 2026 according to SaaStr [23], which puts December 2021 ARR at around $160 million. Against that order of magnitude, $11.7 billion works out at more than sixty times recurring revenue. Any 2021 mark would have been cut down in 2026, AI agents or no AI agents.

Growth deceleration explains a good part of the rest. The same analysis points at the real culprit: tripling revenue over five years and ending up at 20% annual growth on $480 million gets you valued like a cash-flow asset [23].

Semafor, finally, which covered the deal on the day, does not go with the AI explanation. The publication talks about capitulation and a return to valuation reality, without citing agents or seat compression [22]. Flagging that divergence beats burying it.

Airtable illustrates this article’s argument, it does not prove it. The deal establishes that the downward repricing of classic SaaS has moved off the trading screens and into contracts. It does not establish that AI is the sole cause.

The practical consequence, for anyone running processes on Airtable

Bending Spoons has a documented operating pattern on its previous acquisitions: price rises, restricted or scrapped free tiers, headcount cuts. At Evernote, the rise reached 86% [25]. The group, listed on the Nasdaq since 1 July 2026, already owns Evernote, Vimeo, WeTransfer, AOL and Eventbrite, the last two bought in January and March 2026 [20][25].

A leader in that position has a short window, and the question from section 5 with a date attached to it: how many seats billed, for which processes, and which of those would fit in a tool built in-house before closing? Leaving is not necessarily the right answer. But the number is worth knowing before the next pricing amendment lands.

This time, the fall has a signature at the bottom of the page.

Disclaimer. This article is educational and informational. It is neither investment advice nor a recommendation to buy or sell any security. Several figures cited are press estimates, aggregator methodologies that diverge from one another, or non-audited projections, flagged as such in the text. Past performance is no guarantee of future results. For any investment decision, consult a licensed investment adviser.

---

Was your software bill built for 2024, or for a world where producing code costs almost nothing? The answer starts with an AI Maturity Audit.

Sources

Figures in the body of the article are as of the end of July 2026; those in the addendum as of 7 August 2026. Several amounts are press estimates or aggregator methodologies that diverge from one another, flagged as such in the text.

[1] CNBC, “Anthropic updates Claude Cowork tool built to give the average office worker a productivity boost”, 24 February 2026; U.S. News, “Software Stocks Wilted After Claude Cowork’s Debut”: Claude Cowork launched 12 January 2026 (Pro early access), 11 industry-specific modules on 30 January 2026.

[2] Reuters, “Global software stocks hit by Anthropic wake-up call on AI disruption”, 4 February 2026: close to $1 trillion in stock-market value wiped from software and services shares over six weeks.

[3] Forbes (Peter Cohan), “SaaSpocalypse Now: AI Is Disrupting SaaS, But Not All Software Is Doomed”, 6 February 2026: iShares Expanded Tech-Software ETF (IGV) −21%, Salesforce −26%, origin of the term “Saaspocalypse” (a trader), 112→106 SaaS applications per company, 82% reducing their vendor count, per-seat pricing 21%→15% in twelve months, Microsoft/Palantir/CrowdStrike cited by analyst Dan Ives as relatively resilient.

[4] CNBC, “Software stocks rebound as Anthropic announces new partnerships”, 24 February 2026.

[5] Yahoo Finance / 24/7 Wall St., “Salesforce and Adobe Start 2026 With a 5% Plunge”, 6 January 2026: Salesforce −3.5%, Adobe −5.4% as of 6 January 2026.

[6] 24/7 Wall St., “Which Software Stock Has Been the Worst Performer in 2026: Adobe, Salesforce, or ServiceNow?”, 27 April 2026: Adobe and Salesforce both at −31% year to date.

[7] Invezz, “Why are Adobe and Workday stocks falling today?”, 21 July 2026: Salesforce ≈$152 (−43% YTD), Adobe > −35% YTD, Workday −45.5% YTD.

[8] MarketMinute (financialcontent.com), “The Seat-Count Crisis: How AI Agents Triggered the 2026 Software Sell-Off”, 23 February 2026: software sector forward earnings multiple, about 84.1 times at its 2020-2022 peak, fallen to about 22.7 times by March 2026 (market analysis, to be distinguished from official statistical-body data).

[9] TechCrunch, “SaaS in, SaaS out: Here’s what’s driving the SaaSpocalypse”, 1 March 2026: background and precedent, Klarna (late 2024, dropped Salesforce’s CRM for a home-grown system).

[10] TechCrunch, “Lovable says it has hit $500M in annualized revenue, with 1 million new projects a week”, 9 June 2026.

[11] Lovable statistics relayed by the press and specialist aggregators (shipper.now, getpanto.ai): peak of 500,000 projects created or updated on 8 March 2026, daily average then close to 200,000; platform founded in late 2023.

[12] GitHub, Octoverse 2025 report (period 1 September 2024 to 31 August 2025): 121 million new repositories, more than 230 a minute, over 180 million developers on the platform, 36 million new developers in 2025.

[13] SemiAnalysis (Dylan Patel), report relayed by OfficeChai and GIGAZINE, February 2026: Claude Code authoring around 4% of GitHub’s public commits in early February 2026, projected past 20% by the end of 2026. More recent independent tracking suggests a share close to 9 to 10% by mid-2026 (to be treated as a third-party extrapolation, not an official figure published by GitHub).

[14] Apptunix, “iOS Apple App Store Statistics 2026”: roughly 1,249 new iOS apps published a day; 42matters and SQ Magazine, alternative methodologies giving higher ranges (1,800 to 3,600 a day).

[15] Google Play Store statistics 2026 (Apptunix, SQ Magazine): 1,200 to 1,250 new Android apps published a day depending on method; 42matters, up to roughly 2,442 a day under a broader counting methodology.

[16] Stripe statistics 2026 (Chargeflow, RedStag Fulfillment): roughly 1,000 new merchant businesses a day in 2025, with some methodologies running as high as roughly 2,300 a day; 4.8 million active businesses worldwide on the platform in 2026.

[17] Article “Vibe coding is dead: enter UltraCoding” (internal link): 45% of code vulnerable by default (Veracode, 2025-2026).

[18] Article “SaaS & AI: from technical debt to orchestration” (internal link).

[19] Article “Junyr: you are not buying an AI. You are buying the base your AI works on.” (internal link).

Sources for the 7 August 2026 addendum

[20] SEC, Bending Spoons S.p.A., Form 6-K filed 4 August 2026 (Commission File Number 001-43382), signed by Luca Ferrari: acquisition of all outstanding shares of Formagrid Inc (d/b/a Airtable) by Bending Spoons US Inc., $1.285 billion enterprise value entirely in cash, implied equity value of roughly $2.25 billion once net cash is added back, closing expected “later this year”, termination right if the transaction is not consummated by 4 February 2027, and prior transfer of the assets and liabilities of the “Hyperagent” business line to Hyperagent Inc., a Delaware corporation. Bending Spoons press release of the same day: roughly $480 million of annual recurring revenue as of June 2026, growth above 20% year on year, more than 500,000 customer organisations including 80% of the Fortune 100, first acquisition since the 1 July 2026 Nasdaq listing, after AOL (January 2026) and Eventbrite (March 2026). Primary source for this addendum.

[21] TechCrunch, “Bending Spoons to buy Airtable for $1.28B”, 4 August 2026: peak valuation above $11 billion in 2021, roughly $4 billion on the secondary market in early 2026, more than $1.4 billion raised in total; description of the acquirer’s operating pattern (buying at a discount to private marks, trimming staff, streamlining products).

[22] Semafor, “Airtable sale punctures SaaS bubble”, 4 August 2026: “This is what capitulation looks like”, sale at roughly 10% of the peak valuation. Note that this piece explains the deal through valuation correction and mentions neither AI agents, nor seat compression, nor Claude Cowork.

[23] SaaStr (Jason Lemkin), “Airtable Just Sold for $2.25B at 2.7x ARR. It Raised $1.4B and Was Once Worth $11.7B”, August 2026: $11.7 billion peak in December 2021 on a $735 million Series F, $1.4 billion raised across seven rounds, 2.7 times ARR exit multiple on enterprise value, 81% loss of equity value from peak, recurring revenue roughly tripled between 2021 and 2026, median listed B2B software company at around 4.1 times ARR, Hyperagent carved out of the perimeter before the sale. Market analysis written by an investor, to be distinguished from official statistical data; the $160 million December 2021 ARR figure quoted in the addendum is our own calculation from the “tripling” claimed here, not a figure published by Airtable.

[24] Airtable public 2026 price list, concurring aggregator readings (Capterra, Vendr, Adalo): $20 per seat per month on the Team plan billed annually ($24 monthly), $45 on the Business plan ($54 monthly), a seat being any user with editor or commenter rights.

[25] Follow the Money, investigation “WeTransfer owner snaps up apps, then sacks workers and raises prices”; Tech Times, “Bending Spoons IPO Raises $1.68B”, 2 July 2026: post-acquisition operating pattern (price rises, restricted or scrapped free tiers, headcount cuts, legacy apps shut down), 86% price rise at Evernote, Nasdaq listing on 1 July 2026 at $29 per share raising $1.68 billion, Evernote, Vimeo, WeTransfer and AOL portfolio.