Leapfrogging the Pyramid
Why the end of the consulting pyramid is an opening for new talent — not a threat — and how the Alchemy SDLC™ lets newcomers deliver senior-level outcomes on day one
Jamie Campbell | ACC3 International / Alchemist AI Pro™ | July 2026
Introduction: A colleague's challenge
A colleague read The Pendulum Swing [P1] and told me, politely but plainly, that she was not satisfied.
Her objection was not with the economics. That paper argued that the cost of building software has collapsed, that AI-assisted delivery inverts the old buy-versus-build calculus, and that the traditional consulting pyramid, the wide base of junior resources executing billable, repeatable work, is exactly the stratum this shift eliminates. She did not dispute a line of it. Her objection was the mood. Reading it, she said, felt like an elegy: a lament for a disappearing rung and the people standing on it. Find a new angle, she told me.
She was right. The pyramid was a school, the place where analysts became architects, coders became engineers, and testers became the trusted keepers of a release. But it was also a gate. Becoming excellent took years, and those years were rationed: by geography, by hiring cycles, by which engagement you happened to land, by whether a senior person chose to spend their time on you. Plenty of capable people never got through the gate at all.
So the question is not "what happens to junior people when the pyramid stops?" It is "what becomes possible for them when the climb collapses?" My argument is that the collapse of the pyramid is the best thing to happen to new talent in a generation, provided we hand them the right leverage. That leverage is the Alchemy SDLC™: it takes the accumulated method that used to take twenty years to acquire and hands it to a newcomer, so they produce senior-level outcomes on day one and build real judgment far faster than the old path ever allowed.
A note on evidence, before we begin
A word on how to read what follows, borrowed from The Pendulum Swing itself. Some of what supports this paper is historical, dated events, regulatory filings, published reporting. Those claims are marked [H] and cited in full in the references, including every layoff figure in the two tables below. The rest is observable but not yet citable: what a mentor sees across enough cycles, what a delivery team watches happen on real engagements, patterns any practitioner will recognize but no institution has yet written down. Those claims are marked [A], and they are offered for you to test against your own experience rather than to take on my authority. The split is deliberate. The most useful moment in any shift is the one after the evidence is visible and before it is fully citable, which is exactly where the workforce story sits right now. Claims drawn from our own library are marked [P].
Look at the numbers
When I first sketched this argument, one of my mentors stopped me. He has watched more of these cycles than I have, and he has a rule: count before you theorize. Lay the layoffs out plainly, he said, and see what they actually tell you. So I did, and the two views he pushed me toward belong on the page exactly as they landed.
The first view is context. Mass layoffs are not new, and the largest on record dwarf anything a single technology has yet caused on its own. They were the ordinary machinery of economic change, recessions, bankruptcies, corporate break-ups, obsolescence, running at enormous scale. [H1–H10]
Table 1. Ten of the largest workforce reductions on record [H1–H10]
| Company |
Jobs cut (approx.) |
Year |
Industry & cause |
| IBM |
~60,000 |
1993 |
Technology — mainframe decline under new leadership |
| Citigroup |
~50,000 |
2008 |
Banking — credit-crisis response and divestitures |
| Sears Roebuck |
~50,000 |
1993 |
Retail — restructuring; end of the catalog era |
| UPS |
~48,000 |
2025 |
Logistics — network turnaround and automation |
| General Motors |
~47,000 |
2009 |
Automotive — pre-bankruptcy restructuring |
| AT&T |
~40,000 |
1996 |
Telecom — break-up into separate entities |
| Ford |
~35,000 |
2002 |
Automotive — cost-cutting after the 2001 downturn |
| Circuit City |
~34,000 |
2009 |
Electronics retail — liquidation / bankruptcy |
| Boeing |
up to ~30,000 |
2001 |
Aerospace / defense — post-9/11 travel collapse |
| Oracle / Amazon |
~30,000 / ~27,000 |
2026 / 2023 |
Enterprise software / e-commerce |
Figures approximate, as reported; ranked by absolute headcount. Sources, top to bottom, are [H1]–[H10].
The second view is what is different now. When my mentor had me sort recent reductions by industry and keep only the ones where leaders explicitly named AI or automation, a pattern appeared: not one sector, but nearly all of them, and in each case the language pointed at the same kind of work. [H11–H20]
Table 2. Ten significant recent layoffs linked to AI or automation, one per industry [H11–H20]
| Industry |
Company |
Scale |
What was cited |
| Enterprise software / cloud |
Oracle |
~21,000 (~13%) |
Filing: "adoption and deployment of AI" |
| Fintech / payments |
Block |
~4,000 (≈ half) |
AI tools enabling far smaller teams |
| Telecommunications |
BT Group |
up to ~55,000 by 2030 |
2023 plan; AI named in 2025 to go deeper |
| E-commerce / logistics tech |
Amazon |
~14,000 corporate |
Efficiency amid heavy AI investment |
| Logistics / delivery |
UPS |
~48,000 (2025) |
Network restructuring; warehouse automation |
| Banking / financial services |
Standard Chartered |
~7,800 (~15%) by 2030 |
AI to replace "lower-value" work |
| Ride-hailing / mobility |
Uber |
10% of support (2026) |
Simplify the org to "embrace AI" |
| Networking hardware |
Cisco |
~4,000 (2026) |
Realign around silicon, optics, security, AI |
| Cloud / cybersecurity |
Cloudflare |
~1,100 (~20%) |
AI made roles obsolete — "agentic era" |
| Apparel / consumer retail |
Nike |
~775 (2026) |
Automation in distribution centers |
Figures approximate, as reported; several are phased or multi-year plans. Sources, top to bottom, are [H11]–[H20].
Now let me be honest about that second table, in the spirit of the paper it answers. Not every line is a clean case of software doing a named human job. UPS is mostly a network restructuring with automation layered on top; BT's plan predates this wave, and AI has since been named as a reason to cut further; Nike's story is warehouse robotics more than any chatbot. Read the column charitably and skeptically at once and the pattern still holds: across ten distinct industries, leaders are now saying the quiet part in filings and earnings calls, and the roles they name are disproportionately the junior and support functions that once formed the base of the pyramid. Read that Oracle line again [H11]: a company telling its own investors, in writing, that AI cost twenty-one thousand jobs. The base of the pyramid is being repriced with organizations questioning the impact of junior resources versus senior resources assisted with AI, all driven by the gatekeepers (vendor management) who keep pushing the new price close to zero.
But my mentor's conclusion was not despair, and this is the half The Pendulum Swing underplayed. The work most exposed, repeatable, rules-based, moderate-judgment, is exactly the work AI is fastest at compressing. And the same technology is the most powerful reskilling tool ever built. Used deliberately, he argued, AI can compress months of capability-building into weeks, coach a person inside the flow of their real work, and turn disruption into mobility rather than displacement. The new leverage, the skill that makes a person hard to replace, is the ability to direct these tools with judgment: to prompt, evaluate, supervise, and own the result. The barrier and the solvent are the same technology. The only question is whether we hand that leverage to newcomers on purpose, or leave them to find it alone. [A2]
The pyramid was a school — and a gate
Take the school seriously first, because it worked, and because the best of it was never really the work. It was the teaching.
Consider LaTonya. Over more than two decades she became a master business analyst across financial-services institutions including American Express, Charles Schwab, and Bank of America, working every phase of delivery from requirements through UAT and end-user training. But the part that matters most here is what she did with that mastery: she spent years teaching it. She coached junior analysts through the same trials that had shaped her, sitting beside them while a requirement they were sure was obvious blew up in UAT, showing them the question they should have asked, learning across hundreds of those moments what actually turns a novice into a trusted BA. Her expertise was not just personal skill. It was a curriculum, earned through years of trial and error alongside junior resources. [A3]
Now notice who got to attend that school. LaTonya's curriculum reached only the handful of juniors lucky enough to be staffed next to her. Most newcomers never sat beside a LaTonya at all. The teaching existed; it was simply rationed, handed out by proximity and chance to a fortunate few. That is the real gate. It was never only the years the climb demanded. It was that the best teaching, the thing that made the years pay off, reached so few of the people who could have used it. [A1]
Figure 1. The pyramid was a school and a gate — LaTonya's curriculum reached only the few staffed next to her, while hundreds of analysts climbed alone
Which is the question the whole paper turns on: what if LaTonya's curriculum, not just what a great BA knows, but the guided way she taught it, could be handed to anyone, on their first day?
That is precisely what the Alchemy SDLC™ does. The knowledge LaTonya accumulated, not only what a great BA knows, but how she taught it, is now held in Alchemist AI Pro™. A newcomer does not read about her method; they are walked through it. The Socratic elicitation built into the platform is the guided exercise she used to run with juniors, turned into something every user can sit inside. They learn as if they were there beside her: asked the same questions, shown the same traps, corrected at the same moments, except now it reaches everyone, on demand, on real work. [A4]
This is the same insight that runs through our library, turned toward opportunity. Institutional Knowledge Is Leaving Faster Than It Can Be Replaced [P6] shows that the "quiet expertise" of experienced people, their heuristics and judgment calls, can be materialized into consistent artifacts that let less-experienced staff internalize "the decision-making frameworks of departing experts." From Tribal Knowledge to Traceable Requirements [P5] reframes expertise not as a records problem but as a requirements-engineering one: capture the context, elicit the real need, elaborate the rules and exceptions, audit for gaps, export development-ready artifacts. The teacher's method becomes a shared asset instead of a private possession handed out by luck.
Applied to a newcomer, that method is not a safety net to fall back on. It is a launchpad, LaTonya's guided exercise, made repeatable:
- Socratic elicitation walks them through the questions an experienced BA would ask, the very questions LaTonya used to ask her juniors, so the requirement gets interrogated before anything is built, and the newcomer learns why each one matters as they answer it.
- EARS-formatted requirements (Easily Adaptable Requirements Syntax) impose the discipline that removes ambiguity, so a newcomer's specification reads with a veteran's precision.
- Acceptance criteria generated during elaboration set the definition of "done" on day one and, as From Acceptance Criteria to Automated Assurance [P3] describes, export directly into automated test harnesses, so testability is designed in, not rescued later.
- Human review checkpoints let a small number of seniors inspect work in progress, multiplying their judgment across far more delivery than they could personally produce, and giving the newcomer feedback in hours, not years.
This is what leapfrogging looks like. The newcomer is not waiting for permission to do meaningful work; they are producing deployable outcomes immediately, and, because the method makes the reasoning visible, they are being taught as they go, by a curriculum that used to reach only the lucky few. The old apprenticeship built judgment through years of expensive repetition available to a handful of people. This builds it through immediate, guided practice on real work, available to anyone. It is faster, and it is open.
None of this eliminates human judgment. It relocates where a newcomer earns it, out of years of low-stakes grind and straight into the three places where judgment actually forms, with support. Those places are worth naming, because they are where the leapfrog becomes real growth rather than a magic trick.
Requirements gathering: doing on day one what used to take years
The hardest part of requirements work was never the writing. As Mission Readiness Begins with Requirements Readiness [P4] argues, requirements gaps come from fragmented ownership: intent originates with an operator or executive, then passes through program staff, product managers, engineers, and stakeholders, and "no single role owns the entire translation chain from mission intent to development-ready work." The consequences surface late, unconfirmed assumptions, missing criteria, arguments about whether a feature is even done.
AI does not resolve that ambiguity; it amplifies it. Give a model an unclear requirement and it commits confidently to one interpretation and builds it at scale. From Vibe Coding to Governed Delivery [P7] draws the line: a working prototype is not a production system, and "AI can accelerate coding, but it does not resolve the organizational conditions that leave software initiatives misaligned with the mission need." What produces a trustworthy outcome is structured analysis before development.
For a newcomer, this is the leapfrog in its purest form. The judgment LaTonya spent years developing, hearing the requirement underneath the request, sensing when a stakeholder's confidence masks a contradiction, knowing which unstated assumption will detonate in six months, is exactly what the guided elicitation coaches them to do on day one. They will not be as good as LaTonya yet. But they will be running a genuinely senior conversation, capturing it in traceable form, and building the instinct with every session, instead of waiting years to be handed the responsibility.
The final mile: where newcomers accumulate real-world judgment
There is a phase of delivery no amount of generation touches, and From Prompt to Production [P2] names it the final mile: connecting the software to the actual customer environment, identity and access systems, real data sources, external interfaces, deployment pipelines, and the operational realities an AI cannot assume from a prompt.
It is unglamorous and specific: the credential that behaves differently in production, the legacy interface with an undocumented quirk, the integration that works everywhere except the one place that matters. You cannot generate your way through it, because it is defined by a particular organization's particular reality. And that is exactly why it is such fertile ground for a newcomer. The final mile is where the leapfrog stops being about tooling and becomes about a person standing in a real environment, making real calls, accumulating the contextual knowledge that turns a coder into an engineer.
The Alchemy SDLC deliberately keeps humans in this seat: AI-orchestrated development produces the system; people connect it to the world. For new talent, that is not a limitation, it is the growth engine. The years of judgment the pyramid used to build slowly are still built here, but on top of leverage that lets the newcomer arrive at the final mile already productive, rather than spending their first years never getting near it.
UAT and the place where humans do help
If requirements are where delivery starts, user acceptance testing is where accountability lives, and it is the clearest illustration of where humans still do the decisive work.
Automation can verify that the system does what the acceptance criteria say. It cannot, by itself, answer the question UAT actually asks: does this finished system match the original business need? That is a human judgment, and From Prompt to Production [P2] keeps Alchemy UAT as the checkpoint where a person confirms that requirements traceability survived the build.
This is where our most cautionary paper earns its place. Fluent but Fragile [P8] warns that generative AI produces polished, authoritative-sounding output "before users fully understand their source material," that fluent prose "carries assumptions farther before correction becomes expensive," and that a clear-sounding explanation "may hide missing information or contradictions." The antidote is a named human owner who can read and reason with the material, resolve contradictions, and accept accountability, someone for whom, as the paper puts it, "human reasoning remains clear enough to review, correct, maintain, and defend."
For the equalizer argument this is the honest boundary, and, understood properly, the point of the whole thing. The Alchemy SDLC lets a newcomer produce a specification, a build, and a test suite that look senior. UAT is where they step up and say, in effect, I checked this against what we actually needed, and I own it. Learning to stand behind an outcome is how judgment matures. The tool gets a newcomer to that moment on day one instead of year five. It does not stand there for them, and it shouldn't. That accountability is the summit the collapsed climb delivers them to, years earlier than the pyramid ever could.
The equalizer
My colleague was right to push back. Framed as a loss, the end of the pyramid is a sad story about a vanishing rung. Framed honestly, and counted plainly, the way my mentor insisted, it is something better: the removal of a gate that kept expertise slow, expensive, and rationed to the lucky few who got the right seat, and the right teacher, at the right time.
The pyramid was two things, a school and a gate, and AI dissolves both. What it takes away is the years of low-stakes grind that used to be the only on-ramp. What it gives back, through the Alchemy SDLC, is a master's teaching as day-one leverage, LaTonya's guided exercise, available to anyone, plus deliberate human ownership at the three points where judgment is actually earned: gathering the requirement, walking the final mile, standing behind the result at UAT. The same technology that is cutting the base of the pyramid is the fastest tool ever built for turning a capable newcomer into a capable professional.
The industry's new refrain, "hire fewer, demand readiness, use AI to multiply," sounds like a wall in front of new talent. It is only a wall if readiness has to be earned the old, slow way. Supply the readiness as leverage, keep people in the seats where judgment forms, and the same shift flips into its opposite: a genuine equalizer, where the climb to senior-level contribution is measured in weeks of guided work rather than decades of gated apprenticeship. The pyramid stops. The summit opens, to far more people than it ever let up before.
Getting Started with ACC3 International
ACC3 International recommends a focused demonstration with your team or designated staff representatives. Please contact ACC3 International to schedule a demonstration of Alchemist AI Pro™ and assess how it can help turn the collapse of the traditional pyramid into faster, more equitable growth for your own newcomers.
References
The split below follows the convention of The Pendulum Swing. Historical items [H] are dated and citable. Anecdotal items [A] are practitioner observation, offered for the reader to verify against their own view. Primary items [P] are ACC3 / Alchemist AI Pro™ sources.
Historical — dated, verifiable, cited [H]
[H1] IBM — ~60,000 jobs, 1993; mainframe decline under CEO Louis Gerstner. Fortune. https://fortune.com/2015/09/20/biggest-corporate-layoffs
[H2] Citigroup — ~50,000 jobs (some outlets 52,000), 2008; financial-crisis restructuring and divestitures. Fortune. https://fortune.com/2015/09/20/biggest-corporate-layoffs
[H3] Sears Roebuck — ~50,000 jobs, 1993; end of the catalog and store closures. Baltimore Sun. https://www.baltimoresun.com/news/bs-xpm-1993-01-26-1993026112-story.html
[H4] UPS — ~48,000 operations and management roles, 2025; network turnaround and automation. CNBC. https://www.cnbc.com/2025/10/28/ups-earnings-q3-2025.html
[H5] General Motors — ~47,000 jobs, 2009; pre-bankruptcy restructuring and plant closures. Reuters factbox. https://www.globalbankingandfinance.com/factbox-biggest-job-cuts-global-automakers/
[H6] AT&T — ~40,000 jobs (~13%), 1996; break-up into three companies. The Washington Post. https://www.washingtonpost.com/archive/politics/1996/01/03/reorganized-att-to-cut-40000-jobs/13e440c3-81ba-4879-bcab-b790fba7f107/
[H7] Ford — ~35,000 jobs, 2002; "Revitalization Plan" and five plant closures. Fox News. https://www.foxnews.com/story/ford-to-cut-35000-jobs-worldwide-close-five-plants-in-massive-restructuring.amp
[H8] Circuit City — ~34,000 jobs, 2009; liquidation of all stores. The Washington Post. https://www.washingtonpost.com/archive/business/2009/01/17/circuit-city-shutting-down-leaving-34000-out-of-work/81b5fdfa-8e0b-472a-a7bd-1df023bde106/
[H9] Boeing — up to ~30,000 jobs, 2001; commercial-aircraft cuts after the 9/11 travel collapse. CNN. https://edition.cnn.com/2001/US/09/18/gen.boeing.layoffs/
[H10] Oracle — up to ~30,000 jobs, 2026 (CIO.com); Amazon — ~27,000 corporate jobs, 2023 (ABC News / AP). https://www.cio.com/article/4153113/oracle-cuts-up-to-30000-jobs-globally-putting-enterprise-support-and-roadmaps-at-risk.html
[H11] Oracle — ~21,000 jobs (~13%), 2026; regulatory filing cites "adoption and deployment of AI technologies across our operations." Forbes. https://www.forbes.com/sites/maryroeloffs/2026/06/23/ai-cost-21000-jobs-at-oracle-this-year-and-more-layoffs-could-be-coming/
[H12] Block — ~4,000 jobs (~40%, from >10,000 to under 6,000), Feb 2026; CEO Jack Dorsey cites AI tools and smaller, flatter teams. Fortune. https://fortune.com/2026/02/27/block-jack-dorsey-ceo-xyz-stock-square-4000-ai-layoffs/
[H13] BT Group — up to ~55,000 jobs by 2030; plan announced May 2023, with the CEO stating in 2025 that AI could deepen the cuts, largely in customer service. CNN Business; The Register. https://www.cnn.com/2023/05/18/tech/bt-job-cuts/index.html
[H14] Amazon — ~14,000 corporate roles, Oct 2025; framed around efficiency amid heavy AI investment. PBS NewsHour / AP. https://www.pbs.org/newshour/economy/amazon-to-lay-off-14000-corporate-employees-as-spending-on-artificial-intelligence-accelerates
[H15] UPS — ~48,000 roles, 2025; network restructuring with warehouse automation and robotics (see also [H4]). CNBC. https://www.cnbc.com/2025/10/28/ups-earnings-q3-2025.html
[H16] Standard Chartered — ~7,800 roles (>15% of corporate functions) by 2030, announced May 2026; CEO cites replacing "lower-value human capital" with AI. Banking Dive. https://www.bankingdive.com/news/standard-chartered-7800-job-cuts-ai-winters/820627/
[H17] Uber — 10% of customer-service / Community Operations roles, July 2026; memo cites simplifying a "too complex and siloed" org to "embrace AI." Engadget. https://www.engadget.com/2221547/uber-lays-off-10-percent-customer-service-team-ai/
[H18] Cisco — ~4,000 jobs, May 2026; resources realigned around "silicon, optics, security and AI." CFO Dive. https://www.cfodive.com/news/cisco-cut-nearly-4000-jobs-ai-shift-accelerates-layoffs/820302/
[H19] Cloudflare — ~1,100 jobs (~20%), May 2026; CEO says AI made roles obsolete in the "agentic AI era," announced alongside record revenue. TechCrunch. https://techcrunch.com/2026/05/08/cloudflare-says-ai-made-1100-jobs-obsolete-even-as-revenue-hit-a-record-high/
[H20] Nike — ~775 U.S. distribution-center roles, Jan 2026; "advanced technology and automation" in the supply chain. CNBC. https://www.cnbc.com/2026/01/26/nike-to-lay-off-775-employees-at-us-distribution-centers.html
Anecdotal — practitioner observation, offered for verification [A]
[A1] That the consulting pyramid functioned as both a training system and a gate, its years rationed by geography, staffing, and sponsorship, is practitioner observation, offered for the reader to test against their own career.
[A2] The synthesis attributed here to a mentor, that the work most exposed to AI is also the work AI is fastest at compressing, that AI is the most effective reskilling accelerator yet built, and that the result can be mobility rather than displacement, is presented as practitioner observation, not formal research.
[A3] The characterization of LaTonya's expertise as a teachable method, and the claim that such mentorship can be materialized into a guided tool, draws on our own delivery experience and is offered as practitioner observation. Career facts are from her public professional profile.
[A4] The central claim, that a newcomer working inside the Alchemy SDLC can produce senior-level outcomes on day one while building judgment faster than the traditional path allowed, is practitioner observation from our own delivery, offered for verification rather than taken on authority.
ACC3 primary sources [P]
[P1] The Pendulum Swing — Jamie Campbell, Alchemist AI Pro™. https://alchemistaipro.com/library/the-pendulum-swing
[P2] From Prompt to Production: Why AI Software Delivery Needs a Full SDLC. https://alchemistaipro.com/library/from-prompt-to-production
[P3] From Acceptance Criteria to Automated Assurance. https://alchemistaipro.com/library/acceptance-criteria-to-automated-assurance
[P4] Mission Readiness Begins with Requirements Readiness. https://alchemistaipro.com/library/mission-readiness-requirements-readiness
[P5] From Tribal Knowledge to Traceable Requirements. https://alchemistaipro.com/library/fromtribaltotraceable
[P6] Institutional Knowledge Is Leaving Faster Than It Can Be Replaced. https://alchemistaipro.com/library/knowledge-leaving-faster-than-replaced
[P7] From Vibe Coding to Governed Delivery. https://alchemistaipro.com/library/from-vibe-coding-to-governed-delivery
[P8] Fluent but Fragile. https://alchemistaipro.com/library/fluent-but-fragile
