Newsletter · · Ashutosh Agarwal

Cursor Gets Swallowed by SpaceX and the Fight Over Owning Your Intelligence Goes Live - IT Services vs AI - Week of September 5, 2026

IT Services vs AI for the week of September 5, 2026: OpenAI cut off Cursor's model access and SpaceX reportedly bought it, Workday staff built 22,000 custom agents in three weeks without a consultant, and fresh labor data showed the heaviest AI adopters hiring faster, not slower.

IT Services vs AI

Week of September 5, 2026: Cursor Gets Swallowed by SpaceX and the Fight Over Owning Your Intelligence Goes Live

TL;DR

  • The disruptors are eating each other. OpenAI cut off Cursor's access to its models, and Cursor has reportedly been acquired by SpaceX, meaning the most popular AI coding tool is now tied to Elon Musk's Grok. A Factory co-founder says enterprises will now "have a second look" before handing their software work to a model-locked vendor. Meanwhile Cognition (maker of the Devin agent) is reportedly raising at a $46 billion valuation, and Anthropic's coding business alone is doing $800 to $900 million in annual revenue and headed for $1.6 billion by year end. The money and the ambition aimed at replacing developers have never been larger.
  • The clearest disintermediation number of the week came from Workday's earnings. Industry analyst Josh Bersin reported that Workday's own staff built 22,000 custom AI agents in three weeks using a tool anyone can operate: "you don't need to be a software engineer or a very, very technical consultant to build something," in "minutes or an hour as opposed to weeks or months." That is exactly the implementation work integrators like Accenture, Infosys and Wipro get paid to do.
  • But the bear case took a real punch this week. New labor data (via Revelio Labs and a Ramp study) shows the companies adopting AI most aggressively are hiring faster, not slower. One dataset shows a 27% headcount increase among heavy adopters. Tech layoffs are running below their historical average. The "AI collapses the headcount model" thesis now has to explain why the data is pointing the other way, at least for now.

What's New

The five developments most likely to move numbers or the thesis, ranked by how actionable they are.

1. Cursor is now a SpaceX company, and OpenAI has cut its cord. Across four separate podcasts this week, the same story surfaced: OpenAI has restricted Cursor's access to its models, and Cursor has been acquired by SpaceX, which pulls it toward Musk's Grok models. On The Twenty Minute VC (20VC) (Aug 29), Eno Reyes, co-founder of the coding-agent startup Factory and an operator rather than a pundit, spelled out why this matters for anyone selling implementation work: "It's going to be a very hard story to become model independent... when you're attached to a model lab. So they're going to want to push Grok." He added that enterprises "are going to have a second look at the idea of... seeding their software development lifecycle to a provider who is one, likely to be model locked, and two, has an existing... pattern of maybe struggling to operate in these larger and more secure environments." Why it matters: Last week the story was Cursor building its own consulting arm to sell services. This week it got absorbed into a rocket-and-compute conglomerate. The lesson for integrators is that the tool layer is consolidating fast and violently, and enterprise buyers are getting nervous about lock-in, which is exactly the opening a vendor-neutral systems integrator could sell into. The acquisition is reported across podcasts, including the Elon Musk Podcast (Aug 30) and Squawk on the Street (Sep 1), so treat the specifics as podcast-sourced rather than confirmed by a primary filing.

2. "Who is the sovereign of your intelligence?" The model labs are now openly gunning for their own customers' businesses. Also on 20VC (Aug 29), Eno Reyes made the sharpest strategic point of the week. He said "at least two of the largest companies that provide models today have explicitly said, we're going to go after every single one of these industries and businesses that we provide intelligence for." His warning to enterprises: outsourcing all your AI is like a law firm that "outsource[s] every single one of your cases to some other company... come five years later, those other companies can just turn around and screw you over because they know exactly how to do your entire business." He described how the four big coding vendors pitch enterprises: "we're going to build you a... human level AI replacement for labor. And then we're going to sort of Indiana Jones swap this for the people in your business." Why it matters: This is the disintermediation thread coming from the tool side again, and it cuts two ways for integrators. The model labs are a threat to everyone's services revenue, but the fear of being hollowed out by your AI vendor is precisely the anxiety a trusted, neutral integrator exists to soothe.

3. Workday's earnings gave us the cleanest "consultant not required" data point of the week. On The Josh Bersin Company (Aug 31), veteran HR-industry analyst Josh Bersin walked through Workday's latest quarter. The headline for our thesis: Workday deployed its Asana-based agent builder to about 15,000 of its own employees, who built 22,000 custom agents in three weeks. Bersin's line is the one to remember: "now you don't have to be a software engineer or a very, very technical consultant to build something," and you can ship it "in minutes or an hour as opposed to weeks or months." On the money side, he reported that roughly 25% of Workday's new recurring revenue came from AI, with AI contract value "growing at 200% quarter over quarter." He also cited a $600 million AI figure and $100 million of new AI recurring revenue signed in the quarter, though the exact numbers were a little muddled in his readout, so lean on the direction, not the decimal. Why it matters: Configuration and rollout of enterprise software is the bread-and-butter of the Indian IT majors and the Big Four. If customers can self-serve agent-building on top of the platform, a slice of that implementation revenue simply evaporates. This is disintermediation of implementation revenue with a real vendor and a real number attached.

4. Enterprises are increasingly choosing build over buy, and building the capability in-house rather than renting consultants. On The Artificial Intelligence Show (#236, Sep 3), hosts Paul Roetzer and Mike Kaput, who run an AI-education company and so speak as practitioner-pundits, answered a listener asking whether a small firm should hire outside consultants to build its AI agents. Roetzer's answer was blunt: consultants are "a short term fix, honestly... this is going to be so critical to every organization's operational structure... that you have to own it." He described building an internal "Labs" unit that functions "as forward deployed engineers," people who go into each department, find things to automate, and build them. Crucially: "You don't have to go hire software development people. You don't have to go hire AI engineers. They can be business people, knowledge workers with domain expertise." Why it matters: This is the build-versus-buy thread in its purest form. The argument that AI know-how is core IP you must keep inside the company is a direct headwind to the "hire an integrator to transform you" model, and it echoes the forward-deployed-engineer trend we flagged from Microsoft weeks ago.

5. The money aimed at automating software work hit a new high, with Cognition at $46B and Anthropic's coding business near $1B and climbing. On 20VC (Sep 3), the hosts noted Cognition (the company behind the Devin coding agent) is "raising a round at $46 billion reportedly," and that Anthropic's coding business is "currently doing $800 to $900 [million]" and will "end the year at $1.6 billion in ARR." They then debated the size of the prize in a way that should focus every IT-services investor: "one of the ones I've started looking at a lot is just total labor spend and software... You've got about $500 billion a year of US labor spend. And... the big question is what percentage of that converts to AI spend? If it's 10%, it's a $50 billion market." Their bigger point: "people are literally building 100X more software than we were 18 months ago." Why it matters: The bull framing here is Jevons paradox: cheaper software creation means vastly more software gets built, which could mean more integration, more projects, more services. The bear framing is that the $500 billion of US software labor spend the VCs are salivating over is, in large part, the revenue base of the very firms we cover. Both sides are staring at the same pool of money.

The Debate

The bull case: AI grows the services pie and expands margins. This week handed the bulls their best evidence in a while, and it came from data, not vibes. On The Chad & Cheese Podcast (Sep 1), Ben Zweig of Revelio Labs, a labor-data operator and former IBM people-analytics lead, presented findings that cut against the doom narrative: using a Ramp study that tracks companies by AI-token spending, "the most intensive adopters are hiring a lot faster than the... not yet adopters," including among young workers. A separate approach in his own data found "companies that have successfully adopted AI show a 27 percent increase in headcount." His frame is the "productivity J curve": firms invest heavily up front and hire to build, and the payoff comes later. He also pushed back hard on the layoff panic: "layoffs in tech have been actually lower than the historical averages. And most of the layoffs that we see are in manufacturing." The Jevons logic showed up again in an adjacent industry on The Future Is Bright Podcast (Sep 1), where legal-industry analyst Reena SenGupta reported that a Harvey study found law firms are "not really cutting headcount right now. They're reshaping their headcount," with "this explosion of legal engineers," and firms declining to pass price cuts to clients "because we're actually doing more work." And on The Josh Bersin Company (Aug 31), Bersin argued the SaaSpocalypse is reversing: the big platforms are "additive and complementary" with AI, not replaced by it, which keeps the implementation ecosystem around them alive.

"When companies are investing in AI, they're not investing in displacing workers, at least not yet. They're trying to build more." Ben Zweig, Revelio Labs, on The Chad & Cheese Podcast (Sep 1)

The bear case: AI absorbs the billable work and breaks the headcount-growth model. The bears got their ammunition too, and it was more visceral. On Squawk on the Street (Sep 1), Elon Musk, speaking at the G20, predicted AI will be "Stockfish level good" at software within 12 to 18 months, "meaning that it is impossible for a human to compete in writing software with AI... AI will just crush all humans at software." On The Artificial Intelligence Show (#235, Sep 1), the hosts relayed that Bill Gates has turned notably pessimistic on jobs, arguing AI can "substitute for cognition across nearly every industry at once," that many existing jobs "disappear faster than new ones emerge," and floating a "human reserved" category of work plus taxes on AI tokens and robots. And the pricing model itself is under pressure: the same episode noted OpenAI "is now actually experimenting with outcome-based pricing," and both the legal (Future Is Bright) and accounting (The Unique CPA, Sep 1, with Matt Armanino) worlds are visibly shifting from billing by the hour toward value- and outcome-based pricing. The bear's punchline: once the price of professional work is decoupled from hours worked, the entire pyramid of leverage that Accenture, Infosys, TCS and Wipro are built on starts to wobble.

Where it nets out this week: The bear case is directionally strong on a 3 to 5 year horizon but is not yet showing up in the near-term labor data. In fact the data is currently running the other way. That gap between narrative and numbers is itself the trade: if the aggressive AI adopters keep hiring, the incumbents have more runway than the bears think; if the Revelio J curve flattens and headcount rolls over, the re-rate could be swift.

Stocks in Play

Accenture (ACN), not directly discussed on podcasts this week. The only in-window Accenture mention was a passing reference to a decade spent there by a guest on Second in Command (Ep 605, Sep 1), which carried no thesis-relevant content. Read-through applies: the Workday self-service-agent data (22,000 agents in three weeks, no technical consultant needed) and the build-in-house-not-hire-consultants argument on The Artificial Intelligence Show #236 both press directly on Accenture's implementation revenue; the offsetting bull is the Revelio and Harvey "reshaping not cutting, doing more work" evidence. Bull: AI-transformation demand is real and expanding; enterprises frightened of model-lab lock-in want a neutral partner; the SaaS platforms Accenture implements are proving additive rather than replaced. Bear: Self-service agent building and in-house forward-deployed-engineer teams chip at the core implementation franchise; outcome-based pricing threatens the hours-based margin engine. Next catalyst: Accenture's fiscal Q4 print is expected in late September, the single most important data point in our universe. Watch new bookings, the GenAI bookings run-rate, headcount trajectory, and any language on pricing of AI-augmented work. This remains the fundamental follow-through the July bounce needs.

IBM (IBM), not discussed this week. IBM's usual podcast footprint (its own Making Data Simple show, plus Wall Street Week) returned nothing relevant in-window. No update to the AI operating-model moat thread from prior weeks. Bull: The consulting, software and infrastructure bundle positions IBM as an AI integrator of record for cautious enterprises; watsonx and the hybrid-cloud story fit the own-your-intelligence anxiety Eno Reyes described. Bear: Same disintermediation forces; IBM Consulting is not immune to the build-in-house shift. Next catalyst: Consulting bookings and signings, and any quantified GenAI book-of-business update at the next quarterly report.

Infosys (INFY), not discussed this week, the seventh straight week with no direct India IT podcast coverage. Pure read-through. Bull: Cost-takeout and AI-migration demand plays to Infosys's scale; a weaker rupee cushions margins; the reshaping-not-cutting labor pattern, if it holds, protects the pyramid. Bear: Infosys is the most exposed to the linear headcount-to-revenue model that every disruptor this week was trying to break; fresher-hiring and utilization are the pressure points. Next catalyst: Next quarterly result, where the tells are large-deal TCV, headcount net change, utilization, and attrition. Any commentary on GenAI's effect on pricing would be the giveaway.

Wipro (WIT), not discussed this week. Pure read-through, same structural exposure as Infosys with a thinner large-deal cushion. Bull: Turnaround and margin-recovery optionality; AI-led cost-takeout deals could stabilize the book. Bear: Most vulnerable of the four to deal-cycle elongation and to clients internalizing AI work; smallest buffer if bookings slip. Next catalyst: Next quarterly result, and whether the large-deal engine is reaccelerating or stalling.

Read-throughs

  • Enterprise software vendors whose implementations drive services (CRM, NOW, WDAY, SAP). Workday is the standout this week. Josh Bersin (Aug 31) framed it as the poster child for AI being additive to SaaS (about 25% of new recurring revenue from AI, AI contract value up roughly 200% quarter over quarter per his readout), while simultaneously being the poster child for disintermediation (customers self-building 22,000 agents, no technical consultant required). Bersin also flagged Salesforce as "rather schizophrenic on their messaging" and its AI promises as not all proven, a caution for the Agentforce narrative. The double-edged takeaway: platform AI is good for the vendor and bad for the implementer.
  • ERP implementation economics. On Transformation Ground Control (Sep 2), ERP-implementation veteran Eric Kimberling delivered a scathing account of system-integrator incentives: the Big Four and large integrators "make a big share of their revenue from vendor commissions and they depend on vendors for leads," so "their real loyalty is to the vendors," not the client, and the whole ecosystem is being driven "come hell or high water" to maximize the software vendors' annual recurring revenue. His claim that "70% or 80% or more" of these transformation projects are classified as failures is a reminder that the value proposition integrators sell is already under strain before AI even enters the room.
  • Microsoft, GitHub Copilot and the coding-tool layer. No fresh Copilot-specific operator data this week, but the 20VC (Sep 3) market debate and Eno Reyes's enterprise-buyer color are the read-through: OpenAI's Codex is "increasingly referenced in deals... Codex for work," Claude Code is "brought up in every single conversation," and Cursor is now seen mainly as an IDE rather than a full enterprise development strategy. The coding-tool wars are a leading indicator for where in-house developer productivity, and therefore outsourced-developer demand, is heading.
  • Build versus buy and in-house AI. The Artificial Intelligence Show #236 (Sep 3) is the cleanest statement of the own-it-don't-rent-it enterprise mindset: internal Labs teams and forward-deployed engineers drawn from existing business staff, not hired-in consultants.
  • TCS, Cognizant, Capgemini, EPAM, HCL, Tech Mahindra, LTIMindtree. Zero English-podcast coverage again, a persistent structural blind spot rather than a signal that nothing is happening.
  • Adjacent professional services, a live analog. Both legal (Future Is Bright, Sep 1) and accounting (The Unique CPA, Sep 1) are showing the same pattern IT services is heading into: headcount reshaped rather than cut, AI-native new entrants (for example Norm Law AI) taking share from incumbents, and a decisive shift toward outcome and value pricing. Watch these as the leading indicator for how the hours-based model gets repriced.
  • Enterprise adoption temperature check. On Conversations with Institutional Investors (#142, Aug 30), a Fidelity International survey of ASX companies found AI adoption is driving efficiency and some workforce redundancies, but token-cost management remains an unsolved headache, a reminder that enterprise AI economics are still messy, which slows the very disruption the bears fear.

What Changed vs. Last Week

Last week's issue (week of Aug 29) was a genuinely dry week: only two usable episodes cleared the filter, and the one strong source was a Cursor executive describing how Cursor was building its own in-house services and consulting arm. This week the sweep surfaced far more read-through material, and several threads moved.

  • The Cursor story exploded. Last week: Cursor building a boutique consultancy inside itself. This week: Cursor has reportedly been acquired by SpaceX and had its OpenAI model access cut off, a dramatic escalation. The strategic frame flipped from "tool vendor becomes a services competitor" to "tool vendor gets absorbed by a compute giant, raising enterprise trust and lock-in concerns." That lock-in anxiety is a new argument in favor of neutral integrators.
  • Disintermediation confirmed from new, independent sources. Last week it was one Cursor operator. This week the same thread came from three unrelated directions: Workday's self-service agents (no technical consultant needed), Eno Reyes of Factory (model labs going after every industry), and the build-in-house mindset on The Artificial Intelligence Show. Multi-source confirmation strengthens the thesis.
  • Outcome and consumption-based pricing broadened. Last week Cursor flagged the shift to consumption and outcome pricing. This week it showed up as an industry-wide move: OpenAI experimenting with outcome-based pricing, and both legal and accounting firms migrating off the billable hour. The threat to the hours-based integrator model is generalizing.
  • A genuinely new counter-signal for the bulls. This is the first week in a while with hard labor data pushing against the headcount-collapse thesis: Revelio Labs and Ramp showing AI adopters hiring 27% faster, tech layoffs below historical average, and a Harvey study showing law firms reshaping rather than cutting headcount. The "linear headcount model is breaking" bear case now has to reckon with data going the other way in the near term.
  • Still dark: India IT direct coverage (seventh straight week), and dedicated enterprise-agent coverage (Agentforce, ServiceNow, SAP Joule) beyond read-through. The Daybreak and The Ken channel search returned zero again even with a widened window.
  • Unchanged and still the main event: Accenture's fiscal Q4 print in late September remains the real catalyst; nothing this week changes that.

The Line That Sums It Up

Asked on The Neuron (Sep 4) whether advanced AI will "replace consulting firms entirely," the host's answer was a wry no: "all companies need someone other than themselves to blame. So there will always be consultants." Then came the part that actually matters for the model: "I would expect it to be less of them, though... less employees and less jobs there... it's wages that's at stake." That, in one line, is the whole IT-services debate. Not extinction, but fewer people, cheaper hours, and a pyramid that no longer compounds the way it used to.