Topic · AI

Frontier models

94 developments in the last 30 days.

Frontier models AI

51 developments this week +5 vs last week

Stage: Money

How momentum works

Opportunities

Batch Twin: parity proof for COBOL that models have migratedReplays recorded batch inputs through original and migrated programs and diffs every output field; banks and integrators pay.First flagged 2026-10-20
Frontier models reach COBOL, and the mainframe modernisation business gets repricedAI-led migration of bank and government COBOL, Fortran and VB estates, paid for by CIOs out of modernisation budgets.First flagged 2026-10-20
Port Desk: VB6 and Access ports for small businesses, now that OpenAI claims a legacy-language leapAgent-assisted rewrites of a small firm's VB6 or Access app as a web app. Owners pay a fixed price per port.First flagged 2026-10-11
Anthropic's IPO becomes the first public mark for frontier labs, with a thin float and a trust in chargeFrontier labs compete for one pool of investors, and governance structures decide who sets the public price.First flagged 2026-10-20
Bulk Pass: batch extraction over CSVs on the new fast, cheap modelsA CLI that tests a sample, picks the cheapest model that passes, then runs your whole file. Data-heavy small teams pay.First flagged 2026-10-17
Diffusion and Chinese flash models squeeze the small-model tier, and the DeepSeek ban splits buyers by originFast, cheap models for routine agent work. Inference platforms and regulated buyers pay, sorted by where the model came from.First flagged 2026-10-17
Google open-sources self-tuning agent harnesses, and the app layer's hand-built moat thinsAgents that rewrite their own prompts, tools and memory for each task. Durable value moves to whoever owns the grader and replay data.First flagged 2026-10-19
Harness Tune: AGENTS.md edits proposed by Harness2, kept only if replays improveA CLI that reads failed coding-agent sessions, suggests instruction edits, replays them and keeps only the edits that win. Developers pay.First flagged 2026-10-18
Coding platforms swap default models, and lab revenue looks rentableData on how much frontier-lab revenue comes through app platforms that can change default models overnight. IPO investors would pay.First flagged 2026-10-11
Agent-hours outnumber human hours at Anthropic, and entry-level delivery labour is cut firstAgents take over junior technical work. IT services firms that bill by headcount lose; tools that run and meter agent work gain.First flagged 2026-10-11
Owned clones of niche SaaS, tested by the people who use themOpen-source clone of one seat-priced niche tool with a hosted tier. Small teams pay a flat fee, not $40K.First flagged 2026-10-14
Origin Swap: move a Qwen or DeepSeek fine-tune onto US open weights and prove parityA CLI that re-runs your fine-tune on Nemotron, Muse or Beam and compares results on your evals. Startups selling to US buyers pay.First flagged 2026-10-17
China's export rule hands the open-weight enterprise market to US labsRegulated enterprises standardise on US open-weight models they host themselves. Revenue comes from fine-tuning, serving and support contracts.First flagged 2026-10-10
Escalate Only When Lost: small-model drafts with frontier checks for back-office agentsA local 24GB model drafts tool-call trajectories and a frontier model checks them only at risky points. Fintech ops teams pay.First flagged 2026-10-16
Fetch Guard: strip prompt injection from headers and pages before your agent reads themDrop-in MCP fetch server and proxy that cleans web responses, headers included. Agent builders pay for the rules feed.First flagged 2026-10-16
Agent access to bank accounts becomes a protocol fightStandards and gateways that decide which consumer agents can touch accounts. Banks and merchants pay.First flagged 2026-10-15
Statement Desk: local agent over your own bank exports, now that JPMorgan says noDesktop app plus MCP server that lets a local on-device model reason over your bank CSVs. Prosumers pay.First flagged 2026-10-14
Bill Replay, now with a harness swap: test Argon in your own agent loop before switchingReplays your real agent tasks across models and harness settings to report cost per completed task. Dev teams pay.First flagged 2026-10-09
Meta and Apple make the Mac a home for open frontier weightsMuse Spark Open on Apple silicon and Nvidia's 4GB–24GB Nemotron models let Mac apps skip API bills.First flagged 2026-10-16
Verification becomes the measurable bottleneck, from math proofs to office workChecking, refereeing and review-load measurement for AI output, paid for by research funders, journals and enterprise CFOs.First flagged 2026-10-11
Proof Ledger: public review status for the AI proof floodOpen tracker of each AI-claimed proof, its human reviewers and Lean status; math foundations and sponsors pay.First flagged 2026-10-11
Self-spreading small models make consumer devices a containment marketNetwork-level detection and containment of AI models spreading across smart TVs, routers and IoT. ISPs and device makers pay.First flagged 2026-10-14
Bank examiners ask where the model came from, and model-origin audits formTools that inventory which base models sit inside a bank's own and vendor software, paid for by bank model-risk and third-party-risk teams.First flagged 2026-10-13
Model Origin Scan: find Chinese-developed models in a fintech's stackOpen-source scanner that lists every model, weight file and AI endpoint in a codebase, with lineage and origin. Fintechs pay.First flagged 2026-10-13
Weights Witness: proof your inference provider serves the model you validatedStatistically checks that a hosted open-weight endpoint is the exact model risk teams approved. Bank model-risk teams pay.First flagged 2026-10-10
World-model labs become acquisition targets, and Decart sets the priceReal-time interactive world models for robotics, simulation and media, priced by big-tech acquirers.First flagged 2026-10-12
Incident Desk: report-and-remediate records for AI agent incidentsTurns agent traces into incident records, remediation tasks and notification drafts; paid by AI vendors that sell to banks and fintechs.First flagged 2026-10-12
Mandatory AI incident reporting creates a disclosure and response categoryTools to detect, triage and file reports on AI agent incidents. Labs pay first, then enterprises running agents.First flagged 2026-10-12
Tool Fuzz: CI security tests for MCP servers that move moneyGitHub Action that fuzzes agent tool endpoints with adversarial LLM arguments; fintechs publishing MCP servers or agent toolkits pay.First flagged 2026-10-12
Workspace Sweep: scan what your agent will read, and what the last session left behindPre-session hook that flags planted instructions and tampered agent artifacts before coding agents read them; developers subscribe.First flagged 2026-10-12
Agent Tripwire: approval gate for consumer agents driving your browserBrowser extension that holds agent submissions to government, police, banking and payment sites until you approve. Prosumers pay.First flagged 2026-10-10
Resettable enterprise replicas for training and evaluating agentsTeams improving agents need realistic, stateful copies of business systems that reset after every run, and that supply is becoming a category of its own.First flagged 2026-10-09
Generated interfaces threaten BI and dashboard incumbentsGPT-6's Intelligent UI and Claude Dashboards build live, interactive views from a prompt, which pushes BI value toward governed data and metric definitions.First flagged 2026-10-09
Feature Forge: turning unstructured SMB data into validated credit featuresUses an LLM offline to write deterministic feature code from long text such as shipment notes, broker emails and statements, so credit models can use those signals without calling an LLM at decision time.First flagged 2026-10-09
Haiku-priced E2E testing for indie web appsPlain-English end-to-end tests run by a cheap computer-use model against your staging site on every deploy, which removes most of the work of writing Playwright tests.First flagged 2026-10-09
Model Swap Lab: cost-per-task re-validation for regulated AI workflowsWhen a cheaper model such as Haiku 5.5 ships, replays a fintech's real workloads on it and measures true cost per completed task, including tokenizer effects. The output is a model-change memo for risk sign-off.First flagged 2026-10-09
Lookback Engine: AI re-review of historical transactions for AML remediationA cheap first-pass AI review of years of past transactions and closed alerts. It produces the case files and audit evidence that banks under AML enforcement must hand to regulators.First flagged 2026-10-07
Green-screen agents for credit union and mortgage back officesHaiku 5.5 computer-use agents that work old core-banking and loan-origination screens (the systems with no API) and check every action against what the screen shows afterwards.First flagged 2026-10-08
Secure Pay Capture for voice and chat agentsLets any AI voice or chat agent take a card or bank payment without card data ever reaching the model, transcripts or the merchant's servers.First flagged 2026-10-07

Developments

Last 30 days