Newsletter · · Ashutosh Agarwal

The Software Giants Just Hired 12,000 Consultants - IT Services vs AI - Week of July 06, 2026

IT Services vs AI newsletter for the week of July 06, 2026. The week's tell was software companies becoming consultants, as Microsoft ($2.5B), Amazon ($1B) and OpenAI funded armies of embedded implementation experts, even as practitioners described a messy retreat from hourly billing and KeyBank downgraded Salesforce on stalled Agentforce adoption, sharpening the question of whether AI grows the services pie or lets the vendors capture the slice that Accenture, IBM and the Indian majors have owned.

IT Services vs AI

Week of July 06, 2026: The Software Giants Just Hired 12,000 Consultants

TL;DR

The week's biggest tell wasn't a threat to consulting; it was software companies becoming consultants. Microsoft ($2.5B, ~6,000 embedded experts), Amazon ($1B) and OpenAI (a 300,000-certified-consultant program) all committed money to putting humans on-site to make AI actually work. That is the services pie growing and the buyer of services changing at the same time.

The billable hour is being abandoned by the people who invented it. A consulting-software operator said clients now flatly "expect you to be using AI to reduce that effort. So I expect to pay less," and the Wall Street Journal, quoted on-air, called it a "messy retreat from hourly billing," with Deloitte warning labor-based consulting could "shrink dramatically over the next decade."

Salesforce got a fresh, specific negative: KeyBank downgraded it, saying Agentforce is "proof of concept, just not happening." That matters to the integrators because Agentforce-type rollouts were supposed to be their next big implementation payday.

What's new

1. The pivot to services is now funded, not just talked about. For a year the debate has been theoretical: does AI shrink consulting, or does it create a wave of new implementation work? This week the biggest software companies on earth answered with their checkbooks, and the answer was "we want the implementation work ourselves." On the July 7 episode of Business of Tech, host Dave Sobel laid out four moves side by side: Microsoft committing $2.5 billion to a new unit ("Microsoft Frontier Company") that embeds 6,000 engineering and industry experts directly at customer sites; Amazon committing $1 billion to the same embedded-consultant model two days earlier; and OpenAI launching a $150 million partner program with a stated goal of "300,000 certified consultants by the end of the year." His framing of why this is happening is the part that should worry a services PM: Gartner estimates agentic AI could "affect $234 billion of SaaS spending by 2030, roughly 20% of everything businesses spend on software subscriptions," because an agent completes a task across many systems without a human sitting in any one app to do it. The value, in Sobel's words, "never lived in the software. It lived in the job the software helped a person do... Agents just unbolted them." Where does unbolted value go? To whoever owns the outcome, which is exactly what 6,000 embedded consultants and 300,000 certificates are chasing. Business of Tech (Jul 7)

Why it moves the thesis: this is the clearest read yet that the software vendors intend to capture the services layer that ACN, IBM Consulting and the Indian majors have historically owned. It cuts both ways, it validates that AI creates enormous implementation demand (bullish for the pie), while naming new, extremely well-capitalized competitors for it (bearish for who gets the slice). Treat the specific figures as sourced to Gartner, the WSJ and leaked-then-FT-verified OpenAI financials rather than as the podcast's own reporting.

2. The billable hour is in open retreat, and operators are saying so on the record. On the July 10 episode of TECHtonic, Thomas Lah of TSIA (who literally wrote a book called Building Professional Services in 2001) and an executive from consulting-software firm Certinia described the pricing model breaking in real time. The money quote, from the Certinia side: clients "are saying, look, we expect you to be using AI to reduce that effort. So I expect to pay less. So the whole financial construct of effort being valued, being money for us, gets blown up in the AI era." They described the best services organizations scrambling toward outcome- and value-based pricing, borrowing SaaS metrics like net-dollar retention, and facing a collapse in acceptable "time to value," from 12-to-18-month implementations down to three-to-six months. TECHtonic (Jul 10) This lines up with the WSJ line quoted on Business of Tech the same week: a "messy retreat from hourly billing," with Deloitte executives warning that labor-based consulting "could shrink dramatically over the next decade." Business of Tech (Jul 7)

Why it matters: the entire Indian-IT and Big-4-consulting profit engine is people-hours times a margin. If clients now assume AI does the work and refuse to pay for the hours, revenue deflates even if the work volume holds. This is the single most important structural risk for the group, and this week it moved from analyst supposition to practitioners describing it as their lived reality.

3. Salesforce downgraded, the SI implementation tailwind is in doubt. On the July 9 Squawk on the Street, Jim Cramer relayed that KeyBank's Jackson Ader (with Mitch Miller) downgraded Salesforce from buy to hold, saying it was "difficult to find evidence of future upside" and citing "slowing adoption in Agent Force," "agent force, proof of concept, just not happening." CEO Marc Benioff pushed back hard, telling Cramer the call was "wrong" and pointing to "100 references" of Agentforce performing well. The stock is down roughly 40%, the worst performer in the Dow this year. Squawk on the Street (Jul 9)

Why it matters for services: enterprise agent platforms like Agentforce were supposed to be the next great fee pool for systems integrators, someone has to configure and deploy them. If Agentforce adoption is stalling at the proof-of-concept stage, the implementation revenue that Accenture, Cognizant and the Indian majors were counting on gets pushed out with it. It is a demand-side warning light for the SI order book, not just a Salesforce problem.

4. The MSP channel shows what "commoditization" actually looks like on the ground. On the July 9 Business of Tech interview, channel strategist Ryan Morris (25+ years designing go-to-market for vendors and resellers) shared blunt data: about 61% of managed-service providers don't mention AI on their websites at all; only about a quarter of the AI-forward ones have a named, productized AI service; and when a vendor is named, it's Microsoft Copilot roughly 84% of the time. He calls the vague rest "AI washing." His deeper argument is a warning about pricing: AI's consumption model runs opposite to how services firms have always priced. Every per-seat/per-user metric was designed to go up; AI is explicitly bought to make the human-driven units (the thing the provider bills for) go down. As he put it, once a client sees that an agent can do the work of ten users, "why would they continue to pay you for people who are no longer needed?" His counter-hope is the BCG "10-20-70" rule, only 10% of the value is the tech, 20% the data, and 70% the humans, process and change management, which is still services work. Business of Tech (Jul 9)

5. Accenture Song reframed, and AI is already thinning its billable headcount. A sharp Daybreak episode (from The Ken, July 5) put a number on a business most people underrate: Accenture Song generated $20 billion of global revenue in FY25, making it the world's largest advertising agency and driving "nearly a third of Accenture's overall revenue." Its real competitor isn't WPP, it's Deloitte Digital. But the same episode flagged the risk directly: "AI is already cutting into headcount and consequently the number of billable people." The concrete example: Song spent months training an AI model to automate promotional material for Malabar Gold; now that the template exists, "the system generates new banners on its own, which needs much fewer designers." The episode also noted the November 2025 Omnicom-IPG merger created a combined ~$25 billion agency that has now edged past Song. Daybreak (Jul 5)

The debate

Bull: AI grows the pie and the incumbents are the ones holding the shovel. The strongest bull evidence this week came from operators who are getting massive productivity without cutting people, because they have more work than they can staff. On Exchanges (July 9), Citadel's Ken Griffin described an in-house agentic system that reproduces and stress-tests academic finance papers, compressing "6 to 8 weeks" of masters/PhD-level work to "on average 2 to 3 hours." His conclusion is the bull case in one line: "there's no reduction in headcount at Citadel on the back of this breakthrough... I will take every single productivity gain I can get because with the talented people we have, we just have more to go after." Exchanges (Jul 9) On AI to ROI (July 7), AMCS Group's Chief Innovation Officer Evan Schwartz made the same argument explicitly against headcount cuts ("reducing headcount... is a finite game. The best you'll get is zero") in favor of a "person plus AI" model that gets "2, 4, 6, 8 times the value" from experienced staff. He gave a real metric: freeing roughly five hours a day let customer-success reps 2-3x their touches and cut churn from 6% to 3%. AI to ROI (Jul 7) Apply that logic to a services firm: if AI makes each consultant 3x as productive and enterprise AI demand is effectively unlimited, revenue can climb while margins expand, the classic operating-leverage story. And the vendors' own scramble (point 1) proves the implementation demand is real and large.

Bear: AI absorbs the billable hour and the headcount-growth engine breaks. The bear case got more corroboration than the bull this week. The pricing evidence is now coming straight from practitioners: clients "expect you to be using AI to reduce that effort. So I expect to pay less" (TECHtonic, Jul 10), and the WSJ's "messy retreat from hourly billing" with Deloitte's own "shrink dramatically" warning (Business of Tech, Jul 7). Ryan Morris's channel data shows the low end already commoditizing toward Copilot resale, with AI's consumption model actively deleting the units MSPs bill for (Business of Tech, Jul 9). And the disintermediation stories are getting vivid: on Conquer Local (July 7), non-coder Jon Cheney claimed he "built the same software twice. The first time, it cost me $3.2 million in 18 months. The second time, it cost me $400 in one weekend," replicating a $105,000 dev-shop quote in "20 minutes" and reaching a $15,000 contract five days from a standing start. Conquer Local (Jul 7) On Eye On A.I. (July 10), Nexus Black's Kriti Sharma described a small elite team building production-grade industrial software "in a matter of weeks" where it "would have taken... possibly even hundreds of engineers, years," including a whisky-distillery deployment (William Grant) the client expects to save "8.4 million pounds per year" at a single factory. Eye On A.I. (Jul 10) And on the AWS for Software Companies podcast (July 7), Vercel's v0 was shown building a full-stack app and provisioning its own database "from prompt to deployed in under an hour." AWS for Software Companies (Jul 7)

Where I land this week: the bull and bear are describing the same fact from two ends. Griffin and Schwartz keep their people because they are demand-unconstrained, they have "more to go after." A services firm's problem is that its revenue is the hours, not just its cost. If the client captures the productivity as a price cut (the TECHtonic quote) rather than the provider capturing it as volume, the pie can grow while the incumbents' revenue shrinks. The swing factor to watch is pricing power on AI-augmented work, and this week the operator commentary tilted bearish.

Stocks in play

Accenture (ACN), directly discussed.

  • Bull: Song ($20B FY25, ~1/3 of revenue) is a genuine tech-plus-creative asset competing with Deloitte Digital, not legacy ad holdcos (Daybreak, Jul 5). And Accenture is visibly positioning as the partner that helps clients actually implement AI; at Farnborough it pitched itself as the key systems integrator for aerospace/defense AI adoption (Aviation Week's Check 6, Jul 10), and its Dennis Hannigan spent a segment on why enterprise AI adoption stalls without exactly the change-management help Accenture sells (Tech Radio Ireland, Jul 9). That is the "70%" in Morris's 10-20-70 framing, and it's Accenture's to sell.
  • Bear: AI is already reducing Song's billable headcount in production, not in theory ("much fewer designers" after the Malabar Gold automation, Daybreak, Jul 5). The Omnicom-IPG combination ($25B) just overtook Song on scale. And the newly funded vendor push (MS/Amazon/OpenAI, Business of Tech, Jul 7) is a direct bid for the same implementation work.
  • Next catalyst / number to watch: the fiscal Q4 print (Accenture reports in the fall). Watch new bookings and, per last week's issue, whether the managed-services bookings weakness (down double digits year over year) stabilizes and whether the large deals that slipped to FY27 actually convert. Tail risk to flag: a hacker claimed a 35GB Accenture data breach this week (source code and cloud keys); Accenture called it an "isolated matter... remediated... no impact to operations." Unverified, and the same handle overstated a 2024 claim, low probability of being material, but worth monitoring given Accenture touches most of the Fortune 500 (Cybersecurity Today, Jul 10).

IBM was not directly discussed by name this week. Read-through: the funded vendor pivot to embedded consulting (Business of Tech, Jul 7) is the double-edged sword for IBM; it validates the "consulting plus our own AI stack" model IBM has been selling, while adding Microsoft, Amazon and OpenAI as scaled competitors for the same book. Next number to watch: IBM's upcoming Q2 report and, specifically, its generative-AI "book of business" disclosure, the running tally of signed GenAI consulting-plus-software deals is the cleanest gauge of whether IBM is a net winner or loser from this shift.

Infosys (INFY) was not directly discussed by name this week. Read-through only: the billable-hour deflation described by TECHtonic operators (Jul 10) and the WSJ/Deloitte commentary (Business of Tech, Jul 7) bear most directly on the linear-headcount-to-revenue model that Infosys and its peers run. Next number to watch: the quarterly print, large-deal total contract value (bookings), headcount direction, and any explicit comment on AI-related pricing/productivity pass-through to clients.

Wipro (WIT) was not directly discussed by name this week; the same read-through as Infosys applies. Wipro's smaller relative scale makes it more exposed if discretionary IT spend stays soft and mid-size deals get deferred. Next number to watch: bookings/book-to-bill and margin guidance on the next print; any sign that AI is compressing per-deal pricing.

Read-throughs

TCS, Cognizant (CTSH), Capgemini and EPAM were not named in podcasts this week. The relevant read-through is the whole What's new section: outcome-based pricing pressure, the vendors' embedded-consulting land grab, and Agentforce implementation doubt all hit these names' order books. EPAM (pure engineering services) is the most directly in the blast radius of AI coding productivity; Capgemini and Cognizant sit between that and the diversified-SI profile.

Salesforce / Agentforce (CRM): the clearest single-name negative of the week (KeyBank downgrade, "proof of concept, just not happening," Benioff's rebuttal, stock -40% YTD, Squawk, Jul 9). Slower Agentforce adoption means a delayed implementation fee pool for the SIs.

ServiceNow (NOW): only a passing mention. Cramer noted an analyst "sale" call on the stock and CEO Bill McDermott (Squawk, Jul 9).

Microsoft / GitHub Copilot (MSFT): Copilot is now the default, and often the only, productized AI offer in the reseller channel (~84% of AI-forward MSP mentions, Business of Tech, Jul 9), and Microsoft's $2.5B "Frontier" push (Business of Tech, Jul 7) makes it both the biggest arms dealer and a new services competitor. Morris's caveat is the read-through worth holding: 6,000-8,000 Microsoft engineers "barely" cover the Fortune 50, so the vendor land grab realistically reaches the enterprise top tier, not the long tail, which stays contestable for independent integrators.

Build-vs-buy / in-house AI: the disintermediation anecdotes (Cheney's $3.2M-to-$400 rebuild, Nexus Black's weeks-not-years builds, Vercel v0's under-an-hour app) are real but concentrated at the small-business and greenfield-app end. There is still no clean, quantified case this week of a large enterprise firing its systems integrator to build a core platform in-house; the strongest disintermediation evidence remains at the SMB/agency tier, not the Fortune 500 tier the Indian majors and Accenture live on.

What changed vs last week

Last week's issue was the inaugural one; three of its four open threads moved this week, one materially:

Consulting-model shift (was: Rodenhauser/K2 saying output-over-hours billing on a "short runway"): strongly corroborated and escalated. This week practitioners on TECHtonic described clients already demanding lower prices because AI does the work, and the WSJ/Deloitte "messy retreat from hourly billing... shrink dramatically" framing put a timeline on it. This is now the most-supported thread in the coverage.

Disintermediation (was: Devin/Cognition narrative, Chamath's "software factory" idea): escalated from narrative to funded commitments, Microsoft $2.5B, Amazon $1B, OpenAI's 300,000-certificate program. The story is no longer "startups might route around the SIs"; it's "the platform owners are staffing up to take the services revenue themselves."

New this week, an Agentforce warning: last week the enterprise-agent-platform story was mostly a tailwind for SI implementation revenue. KeyBank's downgrade ("proof of concept, just not happening") is the first clear crack in that tailwind, and worth carrying forward.