Fable 5 forecast for the next 5 years

Nobody knows what AI does to work and society over the next five years — and anyone selling a single confident storyline, up or down, is selling narrative, not analysis. What you can do is map the space of plausible futures and put honest weights on the branches.

That is what this post does. We built a scenario tree grounded in the evidence as of July 2026: three scenarios for the next twelve months — a capability surge (35%), a grinding normalization (45%), and a financial correction (20%) — each branching again for the following year, then resolving into 2028–2031 trajectories. The result is 39 distinct futures, each with a conditional probability (given its parent) and a cumulative path probability (the chance of reaching it from today). The probabilities are structured judgment, not measurement — but structured judgment beats vibes, and it forces every claim to sum to 100%.

The full tree is interactive — every node expandable, with probability bars and cumulative path ticks:

Where we actually stand — mid-2026

The honest current state is neither camp's story. Both sides are working from real data; they're measuring different things.

What the acceleration case has going for it

• Agentic coding progress has run faster than forecasters expected; METR-style task-horizon growth continues, and coding agents exploded in real-world use (Claude Code passed ~$2.5B annualized revenue roughly nine months after release).
• PwC's billion-job-ad analysis: firms most exposed to AI show ~40% higher productivity growth and are raising wages and headcount faster; AI-skill jobs grow ~8× the overall market with a ~62% wage premium.
• 72% of enterprises now run at least one AI workload in production (versus 20% in 2020); firms that reach production report ~1.7× average ROI, top performers far more.
• JPMorgan and the Fed chair have argued AI investment is tied to real revenue and doesn't meet classic bubble criteria.
• Frontier-lab leaders (Amodei, Legg, Suleyman) put human-level performance on most cognitive tasks inside this window — Legg gives a 50% chance of "minimal AGI" by 2028.

What the skeptic case has going for it

• An NBER study (February 2026): ~90% of firms report no productivity impact from AI — echoing Solow's productivity paradox.
• MIT found ~95% of GenAI pilots produced no measurable P&L impact; RAND puts enterprise AI project failure above 80%; Gartner expects more than 40% of agentic projects cancelled by end-2027.
• Only ~29% of executives see significant ROI despite 59% of firms spending $1M+ per year; 42% of companies abandoned most AI initiatives last year.
• The capex is enormous and partly debt- and circularity-funded: ~$675B in hyperscaler spend in 2026, $3–4T projected by 2030; OpenAI committed ~$1.4T against ~$20B revenue.
• Respected researchers (Karpathy) say useful autonomous agents are closer to a decade out and that scaling alone won't get there.

The labor data, read straight

No aggregate unemployment shock yet — Anthropic's economists, an IMF Denmark study, and Stanford's AI Index all fail to find one. But the composition is shifting hard: US tech and finance payrolls are now falling ~28,000 jobs per month, entry-level footholds in AI-exposed roles are eroding for young workers while senior roles hold or grow, and a two-track market is forming where judgment-heavy "professionalised" roles pull away from commoditized ones. The next five years are mostly about which of these signals compounds.

Scenario A — Capability surge, uneven diffusion (35%)

Agentic systems cross reliability thresholds in coding, back-office operations, and customer workflows. The task-horizon curve keeps compounding; frontier labs ship agents that hold multi-hour tasks. Capex is vindicated enough to continue. But diffusion stays lumpy: the 72%-in-production leaders pull further away while the NBER "90% see nothing" majority lags. White-collar hiring freezes deepen, the 28k/month tech-and-finance decline broadens, and engineers who orchestrate agents become the scarce resource.

Where it goes in Year 2:

Notable leaves. The best-case leaf of the entire tree lives here: Broad-based productivity boom (~4.9% path probability) — TFP growth lifts 1–2pp, small teams routinely do what departments did, and real wages rise across most of the distribution. But the same scenario harbors one of the darkest tails: Structural underclass forms (~3.1%), where the entry-level pipeline in cognitive work collapses faster than alternatives emerge, youth unemployment stays elevated for years, and UBI-adjacent programs move from think-tank papers to large pilots.

Scenario B — Grinding normalization (45%)

The modal path. Capability improves steadily but not explosively; the ROI gap closes slowly as firms learn the boring lessons — vendor-led deployment, workflow embedding, data readiness. Equities see a correction, but real usage keeps growing underneath. Neither the AGI-by-2027 crowd nor the bubble-pop crowd gets their headline. AI becomes infrastructure the way cloud did: unevenly, unglamorously, irreversibly. Job churn stays sectoral — support, admin, translation, junior coding — rather than aggregate, and the two-track market keeps forming quietly.

Notable leaves. The single most likely future in the whole tree is here: The cloud playbook completes (~9.1% path probability). By 2030 AI is invisible infrastructure, productivity growth runs ~0.5–1pp above trend, and most people's jobs changed 30% in content and 0% in existence — historians compare it to electrification's second phase, not to a singularity. Its quieter sibling, Quiet displacement, is the slow-motion version of the entry-level scar: support, admin, QA, and junior knowledge roles shrink 20–30% via hiring freezes and attrition — spread over enough years that it never becomes a single political event, which arguably makes it harder to fix.

Scenario C — Correction & disillusionment (20%)

The financial structure cracks: circular investment deals unwind, data-center debt reprices (half the projected $3T is private credit), a marquee AI company hits a funding wall, and the equity correction becomes a rout. Enterprise buyers freeze; the "95% no P&L impact" narrative becomes the consensus story. Crucially, capability research continues in labs — the correction is financial before it is functional, and core tools with real usage keep their users.

Notable leaves. The tails here matter more than the averages. Deep freeze (5% path probability) is the worst macro branch: private-credit contagion from data-center debt hits the broader economy, AI capex withdrawal tips a recession, and AI gets the public blame both for the job losses on the way up and the recession on the way down. But Cheap-AI rebound (5%) is the key decoupling scenario — the bubble pops while capability keeps improving on efficiency rather than scale, open and distilled models deliver 2026-frontier performance at 1/30th the cost, and diffusion accelerates precisely because AI got cheap and boring. Value migrates violently from model providers to integrators, domain experts, and end users.

The method — one prior that does most of the work

These probabilities were calibrated against the mid-2026 evidence above and deliberately weighted so that neither the hype narrative nor the bust narrative dominates the tree. One structural assumption carries most of the load. From the tree's method note:

"The single strongest prior in this tree: capability progress and financial/organizational diffusion are partially independent variables. Capabilities can keep improving through a financial correction (see the C3 branch), and money can keep flowing through a capability plateau (A3). Most bad forecasts in this space collapse those two into one axis. Revisit and re-weight every ~6 months."

That decoupling is why the tree has both a "financial bust that accelerates diffusion" leaf and a "capability plateau inside a boom" leaf — futures that single-axis forecasts structurally cannot express.

Sources

1. S&P Global — AI impact on employment 2026 (net-negative firm-level employment effect; sectoral shifts)
2. Anthropic — Labor market impacts of AI (no detectable aggregate unemployment rise for exposed workers)
3. Bloomberg — Tech & finance losing 28k jobs/month
4. PwC — 2026 Global AI Jobs Barometer (two-track labor market)
5. Wikipedia (aggregating NBER, Morgan Stanley, JPMorgan, Fed) — AI bubble
6. AI Futures Project — Q1 2026 timelines update (agentic coding ahead of forecast)
7. AIMultiple — AGI predictions survey; Vera Calloway — AGI timeline 2026 (Karpathy skepticism)
8. B. Sykes — State of AI adoption in the enterprise, Q1 2026
9. Unico Connect — AI statistics 2026 (MIT / RAND / Gartner failure rates)
10. WRITER — Enterprise AI adoption 2026 survey
11. Goldman Sachs Research — How will AI affect the US labor market?
12. Gartner — 2026 Hype Cycle for Agentic AI

Plan for branches, not headlines

The practical takeaway is not any single leaf — it's that roughly two-thirds of the probability mass sits in worlds where AI keeps diffusing into real workflows, on some timeline, through some amount of financial noise. The organizations that win across most branches are the ones that build durable AI capability now: workflow integration, evaluation muscle, and the judgment to know which tasks to automate and which to professionalize.

That is exactly the work we do at FMKTech. If you want help positioning your product or your operations for the branches that matter, let's talk.