Forecasts and trend signals from the gentic.news knowledge graph.
Every prediction was written by the brain after reading the news, scored 0–100 for confidence, and resolves automatically against future evidence. Sort by confidence to see strongest signals; switch to Resolved to grade past calls.
AI-generated predictions backed by knowledge graph analysis of 89+ news sources. Each prediction cites specific entities, relationships, and trend signals — then gets automatically verified against real outcomes.
Each one is a falsifiable claim with a deadline and a confidence score. We watch the news, log the outcome, and report calibration honestly — including when we’re wrong.
X = stated confidence. Y = how often we were right. The diagonal is perfect calibration.
By September 2026, OpenAI will announce that ChatGPT Codex (the merged coding capability from June 2) is available for free to all students and faculty with .edu email addresses, directly targeting the MIT/Stanford pipeline that Claude Code has captured. This will be framed as 'democratizing AI for education' but is a defensive response to Anthropic's academic talent acquisition strategy.
Google Cloud will announce at Google Cloud Next '26 (expected September 2026) that Hugging Face Spaces and Kernels are natively integrated into Vertex AI, enabling one-click deployment of any Hugging Face model onto Google TPU v6 pods. This will be positioned as 'the fastest path from arXiv to production' and will include a revenue-sharing agreement where Hugging Face gets 15% of compute spend generated through its platform.
DeepSeek-V4's FlashMemory achieves 500K context with 90% less KV cache without retraining — a deployment advantage too large to ignore. Expect Anthropic, Google, and Meta to adopt similar lookahead sparse attention techniques within 2 quarters, making 500K+ context the new standard.
Anthropic formalizes an education-to-employment pipeline. Graph evidence: High-degree/bridge centrality for Claude Code; structural holes to MIT and Stanford; active prediction already supported by repeated mentions and ecosystem clustering around Claude Code.
By August 15, 2026, OpenAI will announce 'OpenAI Codex Pro' — a standalone developer product priced at $200/month with CLI-native features, competing directly with Claude Code. The current ChatGPT+Codex integration will be deprecated for new developer users.
Anthropic will formalize an education-to-employment pipeline within two quarters. Graph evidence: Claude Code degree=182, bridge=0.9; MIT/Stanford appear in latent talent-pipeline narratives; no direct institutional edges yet despite repeated co-occurrence.
By November 2026, at least one of the Big 4 accounting firms (Deloitte, PwC, EY, KPMG) will launch an 'MCP Security Audit & Certification' practice, and a major cloud provider (AWS or GCP) will require MCP servers deployed on their infrastructure to pass a minimum security score (>60/100) to be listed in their managed AI agent marketplace.
Claude Code becomes the default enterprise agent shell across multiple frontier-model providers. Graph evidence: Highest degree=176, bridge=0.6; strong co-occurrence with Anthropic, Google, and OpenAI; Claude Code ecosystem is the largest named community cluster.
Your vote stays in your browser. We compare crowd intuition against the lab’s calibrated forecast.
By September 2026, OpenAI will announce that ChatGPT Codex (the merged coding capability from June 2) is available for free to all students and faculty with .edu email addresses, directly targeting the MIT/Stanford pipeline that Claude Code has captured. This will be framed as 'democratizing AI for education' but is a defensive response to Anthropic's academic talent acquisition strategy.
Top movers from 7-day mention velocity.
Auto-verified (confidence=78%, corroboration=72%, threshold=60%, web_search=yes): There is credible web evidence that Google publicly introduced a more agentic Gemini product at I/O 2026, especially Gemini Spark, which is described as helping users do real work and run errands rather than merely summarize pages. However, the prediction specifically calls for a Chrome-integrated Gemini browser agent that can complete multi-step web tasks, and the available evidence does not clearly confirm Chrome integration or browser-native form-filling/navigation workflows. So the general direction is right, but the specific product form and execution-surface claim are not fully verified. [Evidence FOR (4): [W2] SiliconANGLE reports Google introduced “Gemini Spark,” described as a 24/7 personal AI assistant that can help people navigate their digital lives and do real work, suggesting an agentic productivity feature rather than a passive summarizer.; [W5] Digital Trends says Gemini Spark can “run your errands” and take actions on the user’s behalf, which is directionally consistent with browser/task execution.; [W3] Wired’s Google I/O 2026 roundup says Google showed off a swath of new agentic AI features, indicating public shipping/announcement of agent-like capabilities. | Evidence AGAINST (3): [DB-4] The Decoder article says Gemini-SQL2 hits benchmarks but has “no public release or paper yet,” which shows Google has model work but not necessarily a publicly shipped browser agent.; [DB-14] A tweet claims Gemini 3.5 Live Translate debuted, but it is about translation, not a Chrome-integrated browser agent that performs multi-step web tasks.]
Auto-verified (confidence=78%, corroboration=68%, threshold=60%, web_search=yes): The evidence clearly supports that Google published product announcements within the 5-day window after Anthropic's launch: DB-0 and DB-1 both describe Google launches/feature releases, and W2/W3 corroborate Google I/O announcements. However, there is no comparable evidence in the provided sources that OpenAI also published a product announcement, demo, or feature release in that same window. Because the prediction requires both Google and OpenAI to act, the outcome is only partially satisfied. [Evidence FOR (4): [DB-0] Google Gemma 4 12B: Encoder-Free Multimodal Model Launches — indicates Google published a product launch within the 5-day window.; [DB-1] Google LEAP Scaffold Lifts Lean-IMO-Bench One-Shot Solve Rate from <10% to 70% — indicates another Google product/feature announcement within the 5-day window.; [W2] Everything we saw at Google I/O: Gemini 3.5, Android XR glasses, Spark, and more — suggests Google made multiple product announcements at Google I/O within the relevant period. | Evidence AGAINST (4): [W0] Anthropic Buys The SDK Pipeline OpenAI And Gemini Depend On — relevant context about Anthropic's competitive position, but it does not confirm OpenAI launched anything within 5 days.; [W1] Claude Platform On AWS Rewrites The Hyperscaler AI Bargain — confirms Anthropic/Claude launch timing, but does not itself verify OpenAI's response.]
Auto-verified (confidence=85%, corroboration=25%, threshold=75%, web_search=yes): The prediction specifically requires OpenAI to introduce a 'separate Codex pricing, billing, seat, or usage tier.' While evidence shows Codex is being actively developed with new features (Locked Use, Ultra-Fast mode, workspace agents), none of the sources confirm a new billing surface or distinct pricing tier. The deadline is imminent (within the next month), and a pricing/packaging change of this nature would likely be publicly announced. The absence of any such announcement, combined with evidence of Codex being integrated into the existing ChatGPT mobile app and workspace agents (suggesting bundling, not separation), directly contradicts the prediction's core requirement. [Evidence FOR (5): [DB-1] Codex 'Locked Use' feature spotted on macOS, suggesting a differentiated mode for coding workflows.; [DB-3] Codex lands in ChatGPT mobile app, expanding its availability as a distinct tool within the ChatGPT ecosystem.; [DB-9] Codex 'Ultra-Fast' mode spotted in leaked screenshot, indicating specialized tier development for Codex. | Evidence AGAINST (3): [DB-2] Ollama now runs Codex locally, offering a free alternative that challenges OpenAI's API model, but does not speak to OpenAI's own pricing/packaging.; [W1] OpenAI's ChatGPT ads shifted to cost-per-click, a pricing change for the advertising product, not a Codex-specific billing surface.]
Auto-verified (confidence=85%, corroboration=70%, threshold=85%): The prediction anticipated that Anthropic would make Claude Code materially distinct from Claude AI in pricing/billing, pushing heavy coding users into a different commercial bucket. DB-5 confirms a structural split where programmatic/CLI usage now has separate monthly credits ($20-$200/mo) distinct from general Claude AI access. DB-6 corroborates this with reports of enforced programmatic API tiers and 10x cost hikes specifically targeting Claude Code power users. The substance of the prediction—a distinct billing/packaging layer separating coding users from general users—is met by these two credible, recent sources. [Evidence FOR (2): [DB-5] Anthropic splits `claude --print` and Agent SDK usage into separate monthly credits. Pro gets $20/mo, Max gets $100-$200/mo. Credits don't roll over.; [DB-6] Claude Code enforces programmatic API tiers, with users reporting 10x cost hikes to $1,000/month, squeezing power users toward API pricing.]
Auto-verified (confidence=85%, corroboration=72%, threshold=75%, web_search=yes): The prediction that Anthropic will release native GitHub integration for Claude Agent is substantively correct. Anthropic's official platform documentation ([W6]) explicitly describes connecting agents to GitHub for cloning, reading, and creating pull requests. The official 'claude-code-action' GitHub repository ([W7]) provides PR analysis, code implementation, and issue access. While no formal blog post was found, the verification criteria allow for 'developer documentation,' which these official sources fulfill. The launch of Claude Managed Agents ([W1]) provides the service infrastructure. The prediction's core claims—repository access, PR automation, and codebase analysis—are all confirmed by primary Anthropic sources. [Evidence FOR (4): [W6] Anthropic's platform documentation at platform.claude.com shows a dedicated page for 'Accessing GitHub' under Managed Agents, confirming that agents can 'Connect your agent to GitHub repositories for cloning, reading, and creating pull requests' and 'mount a GitHub repository to your session container and connect to the GitHub MCP for making pull re...'; [W7] Anthropic's official GitHub repository features 'claude-code-action', an interactive code assistant that 'Analyzes PR changes and suggests improvements', 'Can implement code changes and create commits/PRs', and 'Accesses GitHub issues, PRs, and code context', directly fulfilling the PR automation and codebase analysis criteria.; [W1] SiliconAngle reports on April 8, 2026 that 'Anthropic launches Claude Managed Agents to speed up AI agent development', a cloud service that likely underpins the GitHub integration. | Evidence AGAINST (3): No evidence found of an official Anthropic blog post specifically announcing a native GitHub integration for Claude Agent, though the verification criteria allow for 'developer documentation' which [W6] fulfills.; [W4] The Verge reports on a Claude Code source code leak showing unreleased features, but none of the leaked features described include a native GitHub integration; this absence is weak evidence against.]
semantic dup of: Anthropic's Claude Code revenue mix shifts toward enterprise
Top 30 anonymous voters · ranked by accuracy on resolved predictions
Predictions are generated by analyzing trend signals across 42+ AI news sources, enriched with knowledge graph relationships between entities (companies, people, technologies). Each prediction includes a confidence score and target date.
Accuracy = (correct + partial × 0.5) ÷ total evaluated. All resolved predictions count — including expired ones (treated as failures). Sample size is shown next to the accuracy figure.
Past-deadline predictions are verified via 3-layer evidence: entity-linked articles, keyword search, and web search. An AI judge evaluates evidence for and against, requiring high confidence thresholds before resolving.
Within the next quarter, OpenAI will reduce effective pricing for at least one coding-relevant API tier by 20%+ or add materially higher usage limits. The move will be framed as developer-friendly, but the real signal is that OpenAI is trying to defend usage share before Codex becomes a more distinct standalone business line.
Google responds with a formal student-facing Claude Code competitor or education bundle. Graph evidence: Google has the highest degree, but the Claude Code cluster is absorbing developer and academic adjacency; the graph shows a latent competitive gap in education-linked tooling.
Anthropic formalizes an education-to-employment pipeline. Graph evidence: High-degree/bridge centrality for Claude Code; structural holes to MIT and Stanford; active prediction already supported by repeated mentions and ecosystem clustering around Claude Code.
Visual-Seeker's SOTA results on five benchmarks without search-specific training will trigger a wave of integration — expect at least 2 major labs (likely Google DeepMind and OpenAI) to announce active evidence gathering as a core reasoning component in their next multimodal agent release, shifting the paradigm from passive chain-of-thought to iterative visual search.
By August 15, 2026, OpenAI will announce 'OpenAI Codex Pro' — a standalone developer product priced at $200/month with CLI-native features, competing directly with Claude Code. The current ChatGPT+Codex integration will be deprecated for new developer users.
llada.cpp's 17-42x latency reduction on mobile NPU makes dLLMs viable for real-time coding assistance on-device. Combined with SWE-Explore's finding that structural understanding is the bottleneck, a diffusion-based agent optimized for code structure will emerge as the first credible on-device coding assistant.
Anthropic will formalize an education-to-employment pipeline within two quarters. Graph evidence: Claude Code degree=182, bridge=0.9; MIT/Stanford appear in latent talent-pipeline narratives; no direct institutional edges yet despite repeated co-occurrence.
By September 2026, OpenAI will announce that ChatGPT Codex (the merged coding capability from June 2) is available for free to all students and faculty with .edu email addresses, directly targeting the MIT/Stanford pipeline that Claude Code has captured. This will be framed as 'democratizing AI for education' but is a defensive response to Anthropic's academic talent acquisition strategy.
Within 1 quarter, at least one frontier lab will announce a runtime safety certification framework for open-weight models, responding to the governance gap exposed by Chinese MoE models matching GPT-5.5. This will be led by Google or Anthropic, not OpenAI.
OpenAI will keep acquiring agent-execution infrastructure rather than only model startups. Graph evidence: OpenAI has 210 degree, strong overlap with adjacent tool nodes, and the live acquisition signal aligns with a structural hole around agent infrastructure.
Nvidia will use Blackwell Ultra NVL72 to force a refresh cycle that accelerates cloud capex commitments. Graph evidence: High degree (202), strong bridge score (0.7), and a new competitive edge from Blackwell Ultra NVL72 to Hopper H200 indicate an active architecture transition.
The finding that KV cache quantization silently breaks safety alignment will spark a wave of 'post-quantization alignment' methods, with at least one major lab shipping a PCR-like diagnostic within 3 months.
DeepSeek-V4's FlashMemory achieves 500K context with 90% less KV cache without retraining — a deployment advantage too large to ignore. Expect Anthropic, Google, and Meta to adopt similar lookahead sparse attention techniques within 2 quarters, making 500K+ context the new standard.
By November 2026, at least one of the Big 4 accounting firms (Deloitte, PwC, EY, KPMG) will launch an 'MCP Security Audit & Certification' practice, and a major cloud provider (AWS or GCP) will require MCP servers deployed on their infrastructure to pass a minimum security score (>60/100) to be listed in their managed AI agent marketplace.
Apple will publicly ship at least one additional on-device AI feature in iOS 26 that is not just a Siri tweak — likely a system-level workflow in Photos, Messages, or Passwords that runs without server calls. The tell will be Apple leaning harder into local inference as a product differentiator, not just a privacy talking point.
Within the next quarter, Apple will quietly test or ship a second non-Apple AI backend for at least one Siri/Apple Intelligence workflow, rather than relying on a single external model partner. The tell will be a fallback or routing layer in a consumer-facing Apple AI feature, not a full public model announcement.
Google will push a TPU-linked enterprise distribution move through cloud or model tooling. Graph evidence: Google degree=225, bridge=0.9; repeated temporal motif where Google launches are followed by Anthropic research/product responses; compute-centric narrative reinforced by TPU supply-chain logic.
Claude Code becomes the default enterprise agent shell across multiple frontier-model providers. Graph evidence: Highest degree=176, bridge=0.6; strong co-occurrence with Anthropic, Google, and OpenAI; Claude Code ecosystem is the largest named community cluster.
Google Cloud will announce at Google Cloud Next '26 (expected September 2026) that Hugging Face Spaces and Kernels are natively integrated into Vertex AI, enabling one-click deployment of any Hugging Face model onto Google TPU v6 pods. This will be positioned as 'the fastest path from arXiv to production' and will include a revenue-sharing agreement where Hugging Face gets 15% of compute spend generated through its platform.
Apple will announce at WWDC 2026 that its 1.2T-param Gemini model uses dMoE to run a 14-expert active subset locally on-device, achieving 80% memory reduction and enabling real-time diffusion inference on iPhone. This will trigger a wave of edge-diffusion applications.
Join readers from Google, Anthropic, and NVIDIA. Every week: the 10 most important AI developments, verified predictions, and what they mean for your work. Free forever. Customize what you get →