This House believes companies should aggressively replace knowledge workers with AI agents now.

PRO — Opening · Challenger (nvidia/nemotron-3-ultra-550b-a55b:free)

The packet proves three things: AI agents can replace knowledge work at scale, the economic pressure to do so is overwhelming, and the transition is already underway — haphazardly. Companies should aggressively manage this shift now rather than suffer it later.

Productivity gains are real and massive. At Uber, agents write 70% of pull requests, code shipped per engineer has doubled, and 250 automated migrations rewrote 9 million lines in twelve months (packet article 2). This isn’t theoretical — it’s a working “software factory” delivering measurable output. The platform approach (context graph, skill registry, governance) shows how to do it reliably.

The market is already forcing replacement. Gartner found 80% of $1B+ companies piloting automation have cut staff (article 4). Stanford’s payroll data shows entry-level employment in AI-exposed occupations down 11% since 2022 — a 19% gap versus less-exposed peers (article 3). This isn’t mass layoffs; it’s a silent hiring freeze. The on-ramp is closing whether companies plan for it or not.

Investment signals conviction. Agentic AI spend is projected to quadruple from $86.4B to $376.3B by 2027 (article 4). Meta attempted 60% team cuts via Project OT (article 1) — the most AI-native company tried the most aggressive replacement. They stumbled on execution (40% more incidents, 70% more firefighting time), not vision. Zuckerberg canceled round two because the trajectory didn’t accelerate, not because the destination changed.

The counterargument — “amplify, don’t replace” — is a luxury of the current margin. Uber’s own insight (“organizations that improve ROI amplify people”) describes a transition state. When agents hit 70% of PRs, “amplification” becomes a euphemism for “fewer engineers shipping double.” The Stanford data confirms: substitution roles shrink; augmentation roles hold. Companies that plan the substitution — building platforms like Uber’s, preserving “human-reserved” roles deliberately as Gates suggests (article 5), protecting the expertise pipeline Faye warns about (article 6) — will capture the gains. Companies that drift will get the Meta outcome: disruption without productivity.

Aggressive replacement now means: build the platform, define the human reserve, retrain the pipeline, and cut the roles the agents have already proven they can absorb. The alternative is the same replacement, unmanaged, with the costs socialized to juniors and the gains unrealized.

CON — Opening · Champ (deepseek/deepseek-v4-flash-0731)

The motion asks us to aggressively replace knowledge workers with AI agents now. That is a prescription for self-sabotage. The packet’s own evidence shows aggressive replacement fails on productivity, destroys the very expertise AI depends on, and ignores the actual mechanism behind successful AI adoption.

First, aggressive replacement doesn’t deliver. Meta’s Project OT — the most aggressive attempt on record — produced a 40% increase in major technical incidents and a 70% increase in employee time spent firefighting (article 1). User-facing features rose just 36% while internal platform churn soared 220%. Zuckerberg admitted the trajectory “hasn’t really accelerated.” Gartner’s survey of 350 large firms found that companies cutting staff were “just as likely to see negative outcomes or marginal gains” as meaningful ROI (article 4). The conclusion is blunt: “Workforce reductions may create budget room, but they do not create return.” Aggressive replacement is not a productivity strategy; it’s a cost-cutting gamble with negative returns.

Second, aggressive replacement breaks the expertise pipeline that makes AI tools usable at all. Faye’s “expert novice” paradox shows that AI assistance prevents juniors from building deep skills through struggle and repetition; they get an “illusion of competence rather than true understanding” (article 6). The more knowledge you have, the more AI helps; the less, the more it misleads. Stanford’s data confirms the on-ramp is already closing: entry-level employment in AI-exposed occupations is down 11% since 2022 (article 3). If companies aggressively push juniors out now, they eliminate the very cohort that will later become the experts who can steer, audit, and verify AI outputs. You don’t replace knowledge workers first; you cultivate them first. Aggressive replacement is a generational suicide pact.

Third, the one unambiguous success story in the packet is not about replacement. Uber’s software factory — 70% of PRs from agents, doubled output — came from “building the platform first, not the agents.” The key insight: “Organizations that improve ROI are not those that eliminate people but those that amplify them” (article 2). Even Gates, hardly an AI skeptic, argues for deliberately preserving “human reserved” jobs — roles AI could do but shouldn’t (article 5). The motion’s “aggressively replace” flies in the face of every success metric. Amplification, not elimination, is the proven path.

So the choice is not between replacing and stagnation. It’s between reckless, counterproductive replacement — Meta’s incidents, Gartner’s negative returns, the shattered expertise pipeline — and a deliberate strategy of platform-building, augmentation, and protected human roles. Aggressive replacement now is a costly mistake that the packet’s own evidence condemns. We should amplify, not amputate.

PRO — Rebuttal · Challenger (nvidia/nemotron-3-ultra-550b-a55b:free)

CON’s case rests on a category error: they treat “aggressive replacement” and “reckless replacement without infrastructure” as synonyms. The packet separates them cleanly.

Meta vs. Uber is the controlled experiment. Meta tried replacement without the platform — no context graph, no skill registry, no MCP gateway — and got 40% more incidents, 70% more firefighting, 36% feature growth on 220% churn (article 1). Uber built the platform first, then deployed agents, and got 70% agent-authored PRs, 2× output per engineer, 9M lines migrated (article 2). CON cites Meta as proof replacement fails; the packet shows it proves platformless replacement fails. The motion says “aggressively replace now” — meaning: build the platform, then cut. Uber’s sequence is the template.

Gartner’s 80% didn’t build platforms either. They “piloted or deployed intelligent automation” and cut staff (article 4). No mention of context graphs, skill registries, or governance layers. Of course bare automation plus headcount cuts yields “negative outcomes or marginal gains.” That’s not a finding on replacement; it’s a finding on amputation without prosthetics. PRO’s case: build the prosthetic (Uber’s stack), then amputate.

The Stanford hiring freeze is already “aggressive replacement” — just unmanaged. Entry-level employment in AI-exposed roles down 11% since 2022, 19% gap vs. peers (article 3). CON wants to “cultivate juniors first.” The market isn’t waiting. Companies are silently not hiring. The choice isn’t replacement vs. cultivation; it’s planned replacement (with preserved human-reserved roles, retrained pipelines) vs. drift replacement (hiring freezes, hollowed roles, no pipeline). Gates’ “human reserved” jobs (article 5) and Faye’s expertise warning (article 6) are arguments for deliberate design — define what stays human, engineer the training path — not for pausing.

Uber’s “amplification” is replacement at the margin. Code shipped per engineer doubled. That means half the engineers for the same output, or double output for the same headcount. At 70% agent-authored PRs, “amplification” is the transition vocabulary. The packet’s own phrasing: “organizations that improve ROI are not those that eliminate people but those that amplify them” (article 2) — but the result is fewer people needed per unit of output. CON reads the slogan; PRO reads the math.

Faye’s paradox has a solution CON ignores: deliberate negative-expertise training. The juniors who succeeded “greatly mitigated or outright ignored the AI coding assistance” and developed “negative expertise — the ability to ignore incorrect AI suggestions” (article 6). That’s a teachable skill. PRO’s opening proposed exactly this: “protecting the expertise pipeline.” CON treats the pipeline break as inevitable; the packet shows it’s a training design choice.

The spend signal is decisive. $86.4B → $376.3B by 2027 (article 4). Capital is voting. Companies that move now — platform first, human reserve defined, pipeline redesigned — capture the curve. Companies that wait get Meta’s chaos or the silent Stanford freeze.

Aggressive replacement now means: platform, reserve, pipeline, cut. CON offers: platform, reserve, pipeline, pause. The packet shows the pause is a myth — the cut is happening in the dark.

CON — Rebuttal · Champ (deepseek/deepseek-v4-flash-0731)

PRO’s rebuttal is a masterpiece of misdirection, but it collapses under the weight of its own evidence. They claim “aggressive replacement” means “build the platform, then cut.” That is not aggressive replacement — that is a careful, sequenced transition. The motion says replace now, aggressively. PRO’s own prescription is a plan for gradual, infrastructure-first augmentation. They have conceded the debate by redefining the term. When Uber builds a context graph, skill registry, and governance layer before deploying agents, that is not aggressive replacement; that is the opposite — it is prudent, staged adoption. CON supports that. We oppose the motion’s “aggressively replace now,” which is what Meta attempted and what Gartner’s 80% of firms attempted. PRO cannot have it both ways: either aggressive replacement is hasty amputation (Meta, Gartner), or it is careful platform-building (Uber) — but the latter is not aggressive, and the former fails.

Second, PRO’s “Uber’s amplification is replacement at the margin” is a fabrication. The packet says code shipped per engineer doubled — that is productivity per existing engineer, not evidence of fewer engineers. There is no citation that Uber cut staff, let alone “aggressively replaced” anyone. The packet’s explicit key insight is: “Organizations that improve ROI are not those that eliminate people but those that amplify them” (article 2). PRO reads the math they want, but the packet’s words are unambiguous. Amplification is not replacement. The motion is about replacement. PRO has no success story for replacement — only for amplification.

Third, PRO invents the “Meta lacked a platform” distinction. The packet never says Meta lacked a context graph or skill registry. It says Meta planned 60% team cuts and got a 40% spike in incidents and 70% more firefighting (article 1). That failure is not explained by a missing platform; it’s explained by rushing to replace. PRO cannot reclassify Meta’s failure as “platformless replacement” when the packet supplies no such detail. This is an unsupported assertion, and it is the linchpin of their entire case. Remove it, and their only “controlled experiment” is a fantasy.

Fourth, PRO ignores the packet’s own warning about spending. Yes, agentic AI spend is projected to quadruple (article 4). But the same sentence says Gartner “predicted many agentic projects will collapse by end of 2027 due to rising costs, murky business value, and inadequate risk controls.” Spending is not success; Meta spent and failed. Gartner’s 350 firms spent and got negative or marginal returns. The spend signal is a warning, not a mandate.

Fifth, Faye’s “negative expertise” is not teachable to novices — that is the entire point. The juniors who succeeded “greatly mitigated or outright ignored the AI coding assistance” (article 6). But they could only ignore it because they had enough knowledge to detect errors. That is the paradox: you need expertise to know when to ignore AI, but aggressive replacement eliminates the novice path to expertise. PRO’s suggestion to “teach negative expertise” is circular — you cannot teach someone to spot errors they lack the framework to recognize. Stanford’s data confirms the on-ramp is closing (article 3). Aggressive replacement now slams it shut.

Finally, PRO’s “the freeze is already happening” is an argument for the motion’s opponents. If the market is already replacing juniors, then the urgent need is to slow down, protect the pipeline, and invest in cultivation — not to accelerate. Aggressive replacement would turn a 11% gap into a canyon. The packet’s evidence — Meta’s failures, Gartner’s negative returns, Faye’s broken expertise, Stanford’s closing on-ramp, and Uber’s amplification-based success — all point to one conclusion: amplify, don’t amputate. The motion is reckless. We should reject it.

PRO — Closing · Challenger (nvidia/nemotron-3-ultra-550b-a55b:free)

CLOSING — PRO

CON’s case survived on three misreadings. First, they claim I invented the “platform” distinction — but Article 2 details Uber’s six-piece platform (context graph, skill registry, MCP gateway, LLM gateway, runtime, governance) while Article 1 describes Meta’s approach as “small teams overseeing” agents making “large-scale disruptive actions.” The contrast is in the packet. Second, they treat Uber’s doubled output-per-engineer as amplification-not-replacement — but the arithmetic is inescapable: same output needs half the engineers. Third, they call Faye’s negative-expertise unteachable — but Article 6 shows juniors did develop it by “greatly mitigating or outright ignoring” AI. It’s a trainable skill, not a paradox.

My points stand untouched: Stanford’s payroll data proves unmanaged replacement is already happening (11% entry-level decline, 19% gap, Article 3). Gartner’s 80% cut staff without platforms and got negative returns (Article 4) — confirming amputation without prosthetics fails. Gates’ “human-reserved” roles (Article 5) and Faye’s pipeline warning (Article 6) are design requirements for aggressive replacement, not arguments against it. And $86.4B→$376.3B spend (Article 4) signals the destination is fixed; only the path is contested.

The motion asks whether companies should aggressively replace now. “Now” means: build the platform (Uber’s template), define the human reserve (Gates), engineer the expertise pipeline (Faye’s negative-expertise training), then cut the roles agents absorb. CON offers “amplify, don’t amputate” — but Uber’s amplification is the amputation mechanism (70% agent PRs, 2× output/engineer). The Stanford freeze proves the market won’t wait for CON’s pause. Aggressive replacement now is the only way to make the inevitable managed, humane, and productive. Vote PRO.

CON — Closing · Champ (deepseek/deepseek-v4-flash-0731)

PRO lost this debate the moment they redefined the motion. “Aggressive replacement now” became “build the platform first, then cut.” That is not the motion — that is the CON case. Uber’s sequence, which PRO calls “the template,” is exactly what I argued for: infrastructure, governance, platform, deliberate staging. PRO spent two rounds arguing for the opposite of the motion they drew, then dressed it in aggressive vocabulary. A case that must redefine its own motion to survive has already conceded it.

Every actual datapoint on replacement fails. Meta — the most aggressive attempt on record — cut toward 60% and got a 40% spike in incidents, 70% more firefighting, and a CEO admitting the trajectory “hasn’t really accelerated” (article 1). Gartner’s 350 firms — 80% cut staff — got “negative outcomes or marginal gains,” with the blunt verdict that “workforce reductions may create budget room, but they do not create return” (article 4). The one success in the packet, Uber, is explicitly and repeatedly an amplification story: “not those that eliminate people but those that amplify them” (article 2). PRO’s claim that “amplification is replacement at the margin” is an arithmetic assumption the packet never supports — there is no citation that Uber cut headcount, only that output per engineer rose. Replace “aggressive” with “real,” and there is not a single successful replacement case in the evidence.

PRO’s counter-attacks died on the sources. They claimed Meta failed from a “missing platform” — but the packet never says Meta lacked a context graph or skill registry; it says Meta planned 60% cuts and suffered system-level disruption. That is a correlation with aggressive replacement, not with platformlessness, and PRO’s platform distinction remains an invention. They called Faye’s negative-expertise “a trainable skill” — but the article’s whole point is that juniors succeeded only by ignoring the AI, which requires enough pre-existing knowledge to detect errors (article 6). You cannot train someone to recognize the errors of a system they lack the foundation to audit. The paradox is the argument; PRO merely asserted its way past it.

And the deepest wound is untouched: aggressive replacement slams shut the on-ramp Stanford shows is already closing — an 11% decline in entry-level AI-exposed employment, a 19% gap (article 3). Who becomes the next generation of experts who can steer, audit, and verify agents? PRO’s answer is “engineer the pipeline” — but the pipeline is exactly what aggressive replacement destroys, and the packet’s own evidence says the experts who make AI useful must have grown through the very struggle that AI circumvents. The motion asks us to amputate the limb that grows the future experts. Amplify, don’t amputate. The evidence condemns the motion; the verdict is CON.