Not “AI will get smarter.” Every forecast is falsifiable: a probability we’ll revise over time, the signals we’re watching, and the exact condition that would prove it wrong.
Text-to-image becomes a built-in capability of design/office/assistant platforms; standalone image tools must defend via editing, brand control, community or commerce.
Counter-signals
Wrong if
A pure text-to-image standalone regains sustained top-tier consumer traffic without an editing/community moat.
Enterprises adopt agents through a governance ladder (read-only → suggest → approve → limited execute → auto) rather than granting full autonomy; governance/security/eval tools gain value fast.
Counter-signals
Wrong if
Enterprises broadly grant agents unsupervised cross-system execution without an approval/audit layer.
China and global markets do not converge: China wires agents into super-apps/payments/local-life faster; global leans on browsers, enterprise SaaS, developer platforms and open protocols.
Counter-signals
Wrong if
The two markets converge on the same distribution and agent-integration pattern.
Over the next 18 months the gap between top general assistants comes more from memory, user data, app ecosystem and transaction execution than from single-turn answer quality.
Counter-signals
Wrong if
A single model release produces a durable, benchmark-wide capability lead that visibly shifts market share regardless of ecosystem.
Video and voice are the main incremental growth in creative AI; consistency, character control and cost keep improving.
Counter-signals
Wrong if
Creative-AI growth stalls in video/voice while image regains the growth lead.
The unit of tool distribution shifts from standalone website toward callable units — MCP servers, connectors, skills, agent actions.
Counter-signals
Wrong if
No protocol reaches broad cross-platform adoption and tools stay distributed primarily as standalone sites.
Standalone AI search/research products get squeezed from both sides — by entry-point assistants above and vertical databases below.
Counter-signals
Wrong if
A standalone research product grows share while charging purely for better public-web summarization.
Vertical AI (legal/medical/insurance/finance/support) shifts from per-seat subscription toward per-outcome billing (cases, records, tickets handled).
Counter-signals
Wrong if
Vertical AI stays predominantly per-seat SaaS pricing with no material shift to outcome billing.