Headline & intro
The scariest AI scenario right now isn’t killer robots or rogue superintelligence. It’s something much more mundane: spreadsheets, procurement systems and software bots quietly hollowing out white‑collar work faster than the economy can adapt.
A new report has just poured fuel on that fear, arguing that autonomous “agentic” AI could trigger a brutal economic feedback loop within two years. Whether you buy that exact timeline or not, the mechanism it describes is uncomfortably plausible.
In this piece, we’ll unpack what’s actually in the scenario, who would be hit first, how it intersects with the “Death of SaaS” thesis, what it means for Europe – and why the real danger isn’t AI itself, but how CEOs and policymakers choose to use it.
The news in brief
According to TechCrunch, analyst firm Citrini Research has published a scenario analysis that imagines the global economy two years from now after rapid deployment of AI agents inside companies.
In their thought experiment, white‑collar unemployment has roughly doubled and the overall stock market has lost more than a third of its value. The mechanism is a self‑reinforcing loop: as AI capabilities improve, firms automate more knowledge work, lay off staff, and become more reliant on in‑house AI agents instead of external contractors or SaaS tools. Lower wage bills boost margins in the short term, but reduced household income suppresses demand, putting new pressure on companies to cut costs further with even more automation.
The scenario focuses on AI agents taking over decisions and workflows that are currently outsourced – from marketing ops and procurement to various optimization and integration services – extending earlier “Death of SaaS” debates. TechCrunch notes that Citrini itself frames this as a scenario, not a firm prediction, yet many commentators find it surprisingly hard to pinpoint exactly where the logic breaks.
Why this matters
The Citrini scenario matters because it attacks the assumption underneath most optimistic AI narratives: that productivity gains will be smoothly absorbed by the economy, as they broadly were with past waves of automation.
This time, the pressure is concentrated on white‑collar, middle‑income roles that sit at the core of consumer demand in advanced economies. Think analysts, coordinators, project managers, procurement specialists, marketing and sales ops, customer support, compliance, finance back‑office – the people who live in email, spreadsheets and SaaS dashboards. Those are exactly the workflows AI agents are being trained to handle.
Who stands to benefit? In the short term:
- Hyperscalers and AI platform providers (Microsoft, Google, Amazon, OpenAI, etc.) that sell the underlying models and infrastructure.
- Large enterprises with the capital, data, and IT maturity to build in‑house agents and rip out parts of their SaaS and contractor spend.
Who loses?
- SaaS vendors and B2B service firms whose value is essentially “we mediate and optimize your business processes.”
- Agencies, BPO providers and consultants who monetize standardized, repeatable tasks.
- White‑collar workers whose tasks can be decomposed into API calls and prompts.
The macro risk isn’t “no more jobs ever”; it’s speed and asymmetry. If cost savings from AI accrue quickly to capital owners while displaced workers struggle to find comparable new roles, aggregate demand falls. You get a deflationary doom loop: lower wages → weaker consumption → more margin pressure → more automation.
That’s the dark twist in Citrini’s argument: AI doesn’t have to be generally intelligent to cause real damage. It only needs to be good enough for CFOs to treat headcount as a variable cost to be optimized away faster than the social and political system can respond.
The bigger picture
Citrini’s scenario slots into several converging trends that have been visible for at least two years.
First, we’re already seeing agent‑like behavior in production: GitHub Copilot and similar tools writing boilerplate code, customer‑service bots handling entire support flows, and sales tools that auto‑generate outreach campaigns and follow‑up sequences. These are still “copilots” more than fully autonomous agents, but the direction of travel is obvious.
Second, the “Death of SaaS” thesis has been circulating in venture circles: the idea that generic horizontal SaaS built on simple CRUD apps is structurally threatened when foundation models plus a bit of glue code can replicate much of the value. Citrini extends this argument: if AI can orchestrate multiple tools, many intermediary layers – integration platforms, optimization engines, even some agencies – start to look vulnerable.
Historically, big automation shocks – the factory robot in manufacturing, offshoring in the 1990s–2000s, or the “China shock” in trade – have been painful but sectoral: they hit blue‑collar workers in specific regions. The white‑collar economy, especially in services, acted as a buffer.
Agentic AI threatens that buffer. For the first time, office workers are as exposed as line workers were in previous waves. The difference is that services account for the bulk of GDP and employment in rich countries; a shock here propagates much more broadly.
At the same time, we shouldn’t ignore the upside scenarios. If AI agents unlock genuine productivity gains and we recycle that surplus into new services, lower working hours, or green transition projects, we could see healthier, more sustainable growth.
So the crucial question isn’t “Will AI agents exist?” – they already do, in early form. It’s who captures the gains, how fast displacement happens, and what institutional buffers we build.
The European angle
For Europe, this debate is particularly acute.
On one hand, the EU’s AI Act, GDPR, DSA and DMA create friction for deploying aggressive, data‑hungry AI agents that make opaque decisions about spending, pricing or hiring. That regulatory drag may slow down unilateral cost‑cutting via AI and give workers, unions and regulators more time to react.
On the other hand, Europe is already wrestling with lower productivity growth and a relative lack of hyperscale AI champions. Many of its digital success stories are exactly the kind of SaaS and B2B process‑optimization firms Citrini thinks are vulnerable. If large clients swap them for in‑house agents on top of US‑based foundation models, Europe could see margin compression in one of its strongest tech segments.
The labour‑market structure is also different. Stronger social safety nets, works councils, and collective bargaining – particularly in countries like Germany, France and the Nordics – may resist rapid white‑collar downsizing. But Southern and Eastern Europe, including major nearshoring hubs, are more exposed: shared‑services centres in Poland, Romania or Portugal are built on exactly the kind of repeatable white‑collar work that AI agents can mimic.
For European policymakers, the question is no longer just how to regulate AI risk, but how to steer AI‑driven productivity so it doesn’t hollow out domestic demand. Tax policy, training programs, incentives for shorter workweeks and co‑determination around automation choices will matter at least as much as model benchmarks.
Looking ahead
Will we really see unemployment doubling and stock markets crashing by a third within two years? That’s unlikely as a central case – economies are messier and slower than clean models suggest.
A more realistic path is a series of rolling “micro‑recessions” by profession. Customer support, content operations, low‑complexity software development, basic finance back‑office, some marketing and procurement tasks – all are already under pressure. Over 2–5 years, we should expect:
- Hiring freezes and attrition instead of headline mass layoffs in many white‑collar functions.
- SaaS consolidation as tools that can’t demonstrate unique data or deep domain value get replaced by AI‑first platforms or internal agents.
- Growing gaps between firms that learn to redesign workflows around AI and those that simply cut headcount and hope.
What should you watch?
- Budget lines: how fast “AI services/agents” grows as a category relative to SaaS and payroll.
- Labour‑share statistics and household income growth versus corporate profits.
- Policy signals: experiments with four‑day weeks, wage insurance, tax changes on automation.
The big risk is political backlash if societies feel AI‑driven gains are hoarded by a narrow slice of firms and investors. The big opportunity is to treat AI as infrastructure – like electrification – and deliberately pair it with policies that expand, rather than shrink, the middle class.
The bottom line
Citrini’s “AI agents destroy the economy” scenario is probably too extreme in its timing, but directionally it captures a real vulnerability: rapid, concentrated gains in white‑collar productivity can destabilize demand if we let them flow almost exclusively to capital.
The technology is not destiny here. The choices CEOs make about reinvestment, and the choices governments make about labour protections and redistribution, will decide whether AI agents trigger a doom loop or a new social contract around work.
The uncomfortable question for every reader: in your organisation, is AI being used to augment people – or to quietly make them optional?
