Newsletter · · Ashutosh Agarwal

IT Services Firms Rewire Pay as AI Deflates the Tools They Sell - IT Services vs AI - Week of August 1, 2026

A synthesis of what podcasts said about AI and IT services for the week of July 25 to August 1, 2026, including Accenture and TCS restructuring pay to slow salary compounding, IBM's CEO arguing the quarter was a deferral not a destruction, and AI coding tools commoditizing fast.

IT Services vs AI

Week of August 1, 2026: The Payslip Shows What the Earnings Slides Would Not

Podcast-driven intelligence on how AI is reshaping IT services and consulting. Coverage window: July 25 to August 1, 2026.

The one line

"IT jobs are being remade," and this week we got the payslip to prove it.

TL;DR

  • The labor model is cracking in plain sight. An Indian business podcast walked through exactly how Accenture and TCS are quietly rewiring pay: half your raise into base salary, half into a one-time lump sum; quarterly bonuses pushed to annual. Moves that cut future salary compounding and make it costlier to quit. This is the headcount-to-revenue decoupling story showing up in real employee contracts, not just on earnings slides. (Daybreak, Jul 26)
  • IBM's CEO went on a media tour to argue the miss was a delay, not a death. Arvind Krishna repeated across two fresh interviews that one-third of the deals that slipped out of Q2 are already back, that 80% of IBM's business is "mission-critical" work that AI won't touch, and that "the vast majority, 70%" of AI's value will spread across a messy mix of models rather than concentrate with one or two giants. (WSJ Tech News Briefing, Jul 28; Everyone Talks To Liz Claman, Jul 25)
  • The tools that displace services work are themselves getting cheaper and easier to swap, fast. New reporting shows enterprises deliberately spreading spend across multiple coding assistants so they can jump ship the moment one raises prices, open-weight and Chinese models halving some companies' AI bills, and Cursor usage falling 17% after it moved to usage-based pricing. Deflation is coming for the software layer that consultants get paid to implement. (The Information's TITV, Jul 28; AI to ROI, Jul 30)

What's new

The five developments that actually move the thesis this week, ranked by how much they should change what you do with a book.

1. The pay-restructuring story is the clearest evidence yet that the old IT-services model is breaking

This is the most important podcast of the week, and it isn't from Wall Street. On Daybreak (The Ken, Jul 26), the reporters laid out something concrete: in mid-May, Accenture employees in Pune were told their annual raise would be split in two, half added to base salary as usual, half paid as a one-time lump sum on June 30. It sounds like a nice cash bonus. It isn't. As the hosts explained, "since only half the raise gets added to the base salary, each future raise is calculated on a smaller number. And that compounds and decreases the salary growth year after year." Base salary also drives gratuity, provident fund, and loan eligibility, so a smaller base quietly shrinks all of those too.

Crucially, this is not just an Accenture move. TCS "recently shifted a part of its quarterly bonus to an annual payout instead," which the hosts described bluntly as a golden handcuff: leave partway through the year and you forfeit the whole bonus, not a slice, so people stay, and every early exit means "one less share of the bonus pool for TCS to pay out." Companies are also grading staff on in-office attendance and how many projects they finish, and penalizing time spent "unassigned to a project."

Why now? Because the business is under real strain, per Sumit Kumar, a VP at research firm Everest Group (an industry analyst, pundit/expert rather than a company insider):

  • Accenture's stock crashed nearly 20% in June after missing earnings, and it posted a 2% decline in new services bookings.
  • The top Indian IT players "have lost nearly 9 lakh crore rupees [roughly ₹9 trillion] in market value over the past five years."
  • The Nifty IT index hit a three-year low in June, after Accenture's weak numbers "signaled a continued downturn."
  • The raises IT staff used to negotiate on a job switch, "about 25-30%," "have dropped to 15-20%."

And the punchline that ties it to our thesis, from the union's read: jobs "are essentially getting cut to reduce costs and divert investment toward building more data centres." The employee benefit line already runs 65-75% of total expenses at companies like TCS, Wipro, and Infosys, so when AI lets you hold headcount flat while revenue grows (or even shrinks), that is where the money moves.

Why it matters: This is the linear-headcount-to-revenue breakdown and the labor-pyramid squeeze captured at the level of an individual payslip. The tell to watch, in the hosts' own words: "Until major players like Accenture, TCS and Wipro start expanding their headcount again, which would signal that AI isn't delivering the expected returns, employees have to assume that more restructuring is coming." Headcount re-acceleration is now a bull signal; continued pay-engineering is a bear signal. (Note: this is analyst and reporter commentary plus one anonymous employee account, not company-confirmed policy detail.)

2. IBM's CEO spent the week arguing the Q2 miss was a timing blip

Krishna's post-earnings tour rolled on into this window with two more interviews, and he stayed relentlessly on message. On Liz Claman's show (operator/insider), he framed the miss narrowly: "The narrow area was large capex deals at very large clients. As we see those begin to come back, then it becomes more of a deferral than a destruction question." His evidence: "one third of all the deals that fell out of the second quarter have already come back." The cause, he says, is that a few dozen Fortune 100 clients rushed to buy AI servers, storage and memory ahead of price hikes, memory and semiconductor costs "going up at 60 percent year over year," and temporarily redirected capex away from IBM. He trimmed the full-year sales growth guide "from what had been more than 5% to a range of 4% to 5%" but held the cash-flow guide, which he pitched as the real tell of conviction.

The more strategically interesting interview was with The Wall Street Journal's Christopher Mims. Krishna's core bet on the shape of the AI market: "I believe that the vast majority, 70%, is going to be in a massive heterogeneous mix, a number of open-weight models, a number of models from other countries... models at the edge." IBM's role is the neutral plumber: "if I go beyond maybe the top 10 or 20 non-tech companies, they are not going to have the in-house skills to talk about what I'll call kernel engineering and model routing. So we can help them... while being neutral or agnostic."

He also gave the clearest defense yet of why AI isn't an existential threat to IBM's revenue mix. Mainframe hardware is only "a quarter of the hardware revenue," and hardware is "just 20% of our total business," but the mission-critical work around it (billing, data, real-time systems) "probably reaches 80%" of IBM, and "every dollar that they spend on mainframe... is $3 to $4 that we get across the other parts of IBM." On the disruption question he was disarmingly direct: "don't you have software that is going to be replaced by an agent? We have very little software. Yeah, that could get disrupted. I acknowledge that."

Why it matters: This is the single richest operator source in the sector, and it directly updates last week's IBM thread. The "deferral not destruction" claim is now measurable, one-third back is the number to hold management to next quarter. The bear read is that "we have very little software that could get disrupted" is a tell that IBM's growth engine is running-the-enterprise annuity, not AI-era expansion, and that a guide cut in the very quarter clients chose AI infrastructure over IBM is, as Claman pushed him on, "the market telling you something a little uncomfortable about IBM's position in the AI stack."

3. The AI coding tools are commoditizing, and enterprises are engineering themselves the freedom to switch

If AI coding assistants are the wedge that displaces services labor, this week's news is that the wedge itself is getting cheaper and more interchangeable. On The Information's TITV, Applied AI reporter Laura Bratton (journalist, informed pundit) shared new field data:

  • Enterprises are deliberately multi-sourcing. "It's absolutely a commonplace narrative that I'm hearing from nearly every CIO and CTO," companies give engineers access to more than one coding agent specifically so they "want to be able to replace one tool with another if one of the AI labs jacks up prices."
  • Pricing power is already diverging. For one dataset (from model-router startup Weave), as Claude Code's cost per user rose, usage still grew, but for Cursor, "even as the median cost per user climbed, usage actually declined 17%" after Cursor moved to usage-based pricing.
  • The "harness" is becoming more important than the model. Open-source harnesses (OpenCode, KiloCode, Kline) now let companies swap a proprietary model for a cheaper or open one for coding work, "you don't necessarily need the latest and greatest model to do coding work."
  • Cursor's enterprise revenue grew "about sevenfold" year-to-date, and usage of its own Composer/Grok models "grew 11-fold," even as some customers (T-Mobile, Neo4j) said they're getting better ROI from OpenAI's Codex.

AI to ROI pushed the cost story further: Coinbase's CEO said publicly the company migrated to Chinese open-weight models (Z.ai's 5.2 and Moonshot's Kimi 2.7) and "cut their AI spending in half, even as their token usage increased dramatically." Cursor released a version of its coding tool built on Kimi as a low-cost alternative. And Andreessen Horowitz "estimated that 80% of their portfolio companies are already using open weight models in AI-based products." Anthropic's response, per the hosts: a cheaper Sonnet 5 model "at a steep discount to Opus... aimed directly at cost-sensitive buyers."

Why it matters: This is the pricing-under-AI-productivity theme and the cannibalization theme at the tool layer. If the AI that automates implementation work keeps getting cheaper and more swappable, two things follow: (a) the ROI of doing that work in-house rather than paying a systems integrator keeps improving, and (b) clients will increasingly expect their integrators to price AI-augmented work down, not up.

4. ServiceNow is the market's referendum on "is enterprise software an AI winner or a loser"

On Chip Stock Investor (Jul 31), analysts Kasey Rossi and Nick (investors/pundits) walked through a name that has become the cleanest expression of the whole debate. ServiceNow is down roughly 50-60% from its highs, "very much in a bear market for software," yet the fundamentals kept compounding: Q2 2026 revenue of $4 billion, up almost 25% year over year, the "20%-plus revenue growth streak" intact, and full-year subscription revenue guided to nearly $16 billion (guidance actually raised).

The bear evidence is in the GAAP line: GAAP operating income fell 55% year over year and GAAP free cash flow fell 10%, as the company front-loads R&D, sales and G&A to launch AI, cybersecurity and CRM products. The bull rebuttal: GAAP EBITDA was still up 9%, the margin hit is the cost of new-product ramp, and a reverse DCF implies the market is pricing in only about 6-7% free-cash-flow-per-share growth over ten years, "an ultra low bar." Management is repositioning ServiceNow as an "AI control tower" to orchestrate human and AI workers, and now claims to be "the eighth largest cybersecurity business in enterprise," chasing "$2 to $3 billion in sales in cybersecurity for 2026, 2027."

Why it matters: ServiceNow is the vendor whose implementations feed a huge amount of SI revenue, so how the market values it is a read on how the market values the whole implementation economy. The setup, 25% top-line growth, but a 55% GAAP operating-income drop and a 60% de-rate, is exactly the "AI tailwind or AI headwind, and the market can't tell which" confusion the sector is stuck in.

5. Workday confirms the fight is over pricing models, not technology

On The Chad & Cheese Podcast (Jul 21), Adam Godson, a Workday executive via its Sana and Paradox acquisitions (operator/insider), said the larger platforms have "varying degrees of readiness" for agents, with "many flows that are fully ready" but others needing "new APIs and... philosophical change in how the data is stored and accessed." His key point cuts to the heart of the "SaaSpocalypse": "the tech's already happening that way... it's more of a revenue model issue." The shift is "being able to charge more for results rather than the subscription," and "if you expect SaaS margins to hold for the next decade... it's clear at this point, that's not likely to happen."

He was pointedly skeptical of the build-your-own-with-AI threat: "you're in a build your own and then maintain it forever... We saw that happen in the mid aughts... very few companies were able to build something of scale and then maintain it."

Why it matters: This reinforces the cross-vendor theme: the technology to build agents is largely here; the disruption is to the pricing model (subscription to outcome and consumption), which compresses the margin expectations baked into every enterprise-software multiple, and, downstream, into the implementation partners who ride those platforms.

The debate

Bull case, AI grows the pie. The steel-man got two fresh data points this week. First, the enterprise-software vendors keep insisting the platform is the safe way to do AI: on The AI in Business Podcast, Salesforce's Sharif Karmally argued a CRM is "the best pre-built layered infrastructure for most AI agents," solving the "context rot" that plagues home-built systems, and pointed out that even Anthropic runs Salesforce as its "single source of truth." Second, the tools argument cuts both ways: cheaper, better coding agents mean more software gets built, and someone has to integrate, govern and secure it. Krishna's whole IBM pitch is this: a heterogeneous, messy AI world that most enterprises can't navigate alone needs a neutral integrator, and every mainframe dollar still drags $3 to $4 of software and services behind it. If agents proliferate, the integration surface area explodes.

Bear case, AI absorbs the billable hours and breaks the model. The evidence this week was heavier on this side. Accenture posted a 2% drop in new services bookings and is re-engineering pay to defend margins, not fund growth; you don't restructure raises to slow compounding when demand is booming. The coding tools that do implementation work are getting cheaper and swappable, pushing build-vs-buy math toward "build," and teaching clients to expect AI-augmented work to cost less. The pricing model everyone is racing toward, outcome- and consumption-based, directly attacks the headcount-times-bill-rate engine that IT services was built on. And the union's framing on Daybreak is the cleanest statement of the bear thesis: jobs are being cut to redirect cash into data centers, and nobody is re-expanding headcount yet.

The swing factor: headcount. Both sides agree on the test. If Accenture, TCS, Wipro and Infosys start growing net heads again, the pie is growing and AI is a demand tailwind. As long as they keep freezing hiring and re-engineering comp, the bear model, AI quietly deflating revenue per head and eventually revenue, is winning the argument.

Stocks in play

Accenture (ACN)

  • Bull: Still a machine of scale. Former CMD Sanjay Jain noted on The Brand Called You (Jul 27) that Accenture is now "800,000 strong," with "half of them, that's 400,000, in the capability centers" across Poland, Brazil, Spain and elsewhere, a global talent-agility engine ("the right talent at the right place at the right time") that is hard to replicate. Salesforce's SMB story implicitly needs integrators to stand up agentic CRM at enterprise scale.
  • Bear: The Daybreak episode is a straight negative: a ~20% June stock crash, a 2% decline in new services bookings, and pay-restructuring that reads as cost defense. Last week's valuation bear (Sven Carlin) argued reported 8% EPS growth is closer to 4% ex-charges and true growth is 1-3%; nothing this week contradicts that.
  • Next catalyst / number to watch: New-bookings growth and book-to-bill on the next print, and any sign of net headcount re-expansion.

IBM (IBM)

  • Bull: 80% of the business is "mission-critical" annuity and OPEX work AI won't easily displace; one-third of slipped deals already back; the $3 to $4 mainframe multiplier; 85% of mainframe clients still growing capacity; the "neutral plumber for a 70%-heterogeneous AI world" positioning; plus quantum optionality (HRL Labs acquisition; Krishna now pegs quantum as "a trillion-dollar impact... by the end of the next decade").
  • Bear: Full-year guide cut from "more than 5%" to "4% to 5%"; clients literally chose AI infrastructure over IBM in late June; Krishna's own admission that some software "could get disrupted"; and pundits flagging a sharp mainframe-cycle drop. One host on The Rundown (Jul 23, pundit, unverified) put Q2 revenue at +1% to $17.2B, adjusted EPS at $2.93 (a slight miss), and mainframe sales down 42%.
  • Next catalyst / number to watch: Whether the remaining two-thirds of slipped deals come back over the next one-to-two quarters, and consulting bookings (guided only 1-3%, "not double-digit growers").

Infosys (INFY)

  • Read-through only. Named alongside Accenture, TCS and Wipro on Daybreak as one of the majors whose employee-benefit costs run 65-75% of total expenses and whose headcount trajectory is the key AI-disruption tell. The Nifty IT three-year low and the ~₹9 trillion five-year market-cap loss are the sector backdrop.
  • Next catalyst / number to watch: Net headcount adds and fresher-hiring plans; any large-deal TCV commentary.

Wipro (WIT)

  • Read-through only. Same Daybreak framing: Wipro is explicitly on the "until major players start expanding headcount again" watch list, and carries the same 65-75% employee-cost structure that makes it maximally exposed to the labor-model reset. Treat this as low-confidence.
  • Next catalyst / number to watch: Headcount and utilization on the next print.

Read-throughs

  • TCS: Covered indirectly and negatively, the quarterly-to-annual bonus shift and "golden handcuff" retention mechanics are a margin-and-attrition play, not a growth signal (Daybreak, Jul 26).
  • Salesforce / Agentforce (CRM): Two useful angles. On The AI in Business Podcast, Salesforce's Sharif Karmally described customer Asymbol's agent "Teddy," which "works over a thousand leads a week... does what would take five people" while paired with a human, a concrete productivity-multiple data point and a "CRM as pre-built agent infrastructure" pitch. But on the CPQ Podcast (Jul 26), industry expert Javid Jafar said the Agentforce and Revenue-Cloud "end of sale" of the old CPQ "created a lot of confusion," and estimated only ~30% of customers are really using AI in CPQ in production; "70% are still working through it." He also flagged a subtle SI dynamic: clients now ask integrators to "create those AI-enabled skills... on my coding platform of choice, be it the Cursors of the world or Claude Code" so they can self-maintain, buying the integrator's knowledge as reusable agent skills rather than its hours.
  • ServiceNow (NOW): Add an operator note to the section above: on The Customer Success Pro Podcast (Jul 29), a ServiceNow exec framed AI agents starting with low-risk incident triage and interfacing "across Workday, Oracle, ERP," the platform-as-orchestrator pitch.
  • Microsoft / GitHub Copilot (MSFT): One striking penetration stat from The Information's TITV: "there's 20 million paid Copilot seats, which is a really small fraction of the 450 million total paying Office users." Even the flagship coding assistant is early in enterprise penetration.
  • Build-vs-buy / in-house AI: The recurring counter to disintermediation, argued most sharply by Workday's Godson: building your own is "build your own and then maintain it forever," and the mid-2000s wave of DIY software mostly failed to scale (The Chad & Cheese Podcast, Jul 21). The bull case for both platforms and integrators rests heavily on this "maintenance is the real cost" argument holding.

What changed vs last week

  • IBM, reinforced, not contradicted. Last week we logged the Jul 23 earnings-cluster interviews. This week Krishna's tour extended to Liz Claman (Jul 25) and the WSJ (Jul 28), and the story held together: the "one-third of slipped deals already back" figure recurs, "deferral not destruction" recurs, and the "70% heterogeneous AI market" framing got its fullest airing yet. New this week: the quantum estimate was explicitly raised to a "trillion-dollar impact by the end of the next decade," and the HRL Labs (ex-GM/Boeing) acquisition was tied to it. Higher confidence in "deferral," but the proof is still the next print.
  • Agentforce pricing and POC-stall, corroborated again from a new angle. Last week's thread was Salesforce's shift to $2-per-resolution outcome pricing to unstick stalled deals. This week the CPQ Podcast independently confirmed the demand-side confusion (the old-CPQ end-of-sale "created a lot of confusion," only ~30% in production) and gave a genuine adoption bright spot (Asymbol's "Teddy"). The picture is consistent: real product traction, real deal-cycle friction from pricing and packaging changes.
  • ServiceNow, the market verdict landed. Last week we had the CPO (Amit Zavery) touting a $1.5B AI business and a consumption-pricing shift. This week we got the stock's answer: down ~60%, GAAP operating income down 55% year over year, and a pivot narrative into cybersecurity. New information, same underlying "winner or loser?" ambiguity.
  • Workday now covered. The Chad & Cheese interview supplies the operator voice on agent readiness and outcome pricing.