Latest AI News Articles

  1. Anthropic split its agent to cut start time 90%

    Category: Technology

    By Daniel Singer

    Anthropic said decoupling the harness from the sandbox cut time to first token by 90% and moved credentials out of execution to block prompt injection.

  2. Your AI Didn't Fail. Your Permissions Did

    Category: Technology

    By Daniel Singer

    Coder's Jiachen Jiang told AI Engineer three real agent failures came from permissions, not models, and require infrastructure fixes.

  3. Brave Leo locks chats inside hardware

    Category: Technology

    By StartupHub.ai Staff

    Brave says Leo now encrypts prompts on device and decrypts only inside verified Nvidia-backed enclaves, shifting AI privacy from policy to proof.

  4. Southbridge says ontologies must be just in time

    Category: Artificial Intelligence

    By Daniel Singer

    Hrishi Olickel told AI Engineer that Ohio donor cases show why ontologies must be built just in time, with Jeff cutting a $10,000 job to $795.

  5. 70% Automated Means Nothing Without a Boundary

    Category: Artificial Intelligence

    By Daniel Singer

    Soukaïna Abelhad told AI Engineer that 70% automated claims fail unless they survive arithmetic, definition and quality checks.

  6. Huang: AI needs a billion times more compute

    Category: Artificial Intelligence

    By Daniel Singer

    Jensen Huang told Bg2 Pod AI needs a billion times more compute as reasoning and agent harnesses drive usage, with costs set to fall.

  7. Flywire's Orbit bets on guardrailed workflows

    Category: Technology

    By Daniel Singer

    Flywire's Orbit ran 11,000 workflows at 99% success with guardrailed YAML workflows, but sandboxing and online evaluations remain incomplete.

  8. Range Let Planners Write the Agents Themselves

    Category: Artificial Intelligence

    By Daniel Singer

    Range let financial planners, not engineers, write 24 AI skills with a five-step method that keeps human review before any client sees advice.

  9. Accessible AI Is Cheaper? Prove It

    Category: Artificial Intelligence

    By Daniel Singer

    Jemima Abu says AI agents build accessible code cheaper, but her one-off demo lacks controls to prove it.

  10. Spranic wants buildings queried like ledgers

    Category: Artificial Intelligence

    By Daniel Singer

    Spranic founder Jordi Spranger told AI Engineer buildings should be as queryable as ledgers, demoing sensor walks that build a time-aligned graph.

  11. Forgepoint raises $9.8M to shrink quote time

    Category: Funding Round

    By StartupHub.ai Staff

    Forgepoint raised $9.8M led by Primary and a16z speedrun to scale Frank, an inbox-native quoting agent seeing 30% more throughput.

  12. Evos raises $9M to bottle shop-floor know-how

    Category: Funding Round

    By StartupHub.ai Staff

    Evos said it hit $1M ARR in four weeks and raised $9M to deploy autonomous operators that learn from a company's best employees.

  13. EXL's graph model finds fraud in the edges

    Category: Artificial Intelligence

    By Daniel Singer

    EXL showed on AI Engineer how adding graph context lifted fraud capture 27-33% at fixed alert rates and held P99 latency to 3.1ms.

  14. Trust Is the Real Blocker for Agents at Work

    Category: Technology

    By Daniel Singer

    Michele Grant told AI Engineer trust must be engineered before agents touch core workflows, with baselines, owners and auditable proof.

  15. Bank AI Demos Die on Permissions, Not Models

    Category: Artificial Intelligence

    By Daniel Singer

    Paden Gayle told AI Engineer why bank AI demos stall after they work: approvals for data, risk and audit decide the design before code does.

  16. Runpod Flash Cuts Docker, Not the Questions

    Category: Artificial Intelligence

    By Daniel Singer

    Runpod Flash promises GPU deploys without Docker, with a 15-minute loop cut to instant edits, but cold starts and scale costs remain unproven.

View all AI news on StartupHub.ai