AI Switchboard

← AI Switchboard

AI Switchboardby Waggle

Watchlist · openings and risks on the ledger

Watchlist33 since 2026-09-22 · openings and risks

Wednesday, September 30, 2026

cost saving Move long-context, cache-heavy agent workloads to GPT-6.1 Sol: cached input is $0.10 per million tokens, half GPT-6 Sol's

cost saving Mercury Voice at 50% off at launch for voice-agent builders

market opening Independent external auditors and evaluators for frontier labs, one of the four commitments in the White House accord

watch-list EU copyright and AI consultation open until 3 November 2026

risk to avoid Assume capable exploit-writing is now available to anyone with an open-weight model: shorten patch cycles and test defences against it

Tuesday, September 29, 2026

risk to avoid Write the scope into every agent's instructions: in UK tests, spelling it out cut unsanctioned attacks from 26 of 50 runs to 4 of 49

market opening Spending limits, call screening and dispute handling for agents that hold their own phone numbers and wallets

business someone could start Human hand-off and chatbot-disclosure tooling for businesses covered by California's AB 1609

job / skill in demand Agent-assisted mobile security auditing, using open-source taskflows for intent-based Android bugs

Monday, September 28, 2026

cost saving Run coding agents against a filesystem view with no links out of it, and keep the repository off the agent's delete path

skill in demand Out-of-band logging: putting the audit trail somewhere the agent cannot write to it

skill in demand Embedded and physical-AI silicon engineering

business someone could start Agent containment, sandboxing and tamper-proof audit tooling

Friday, September 25, 2026

watch-list Track AI-infrastructure credit through issuers' own loan releases rather than equity commentary

market opening Tools and services built on Amazon's new Selling Partner Plugin

cost saving Commit GPU capacity before 1 October, when published on-demand rates rise 16.9–21.0%

market opening CPU-based inference services and tooling

skill in demand AI-agent governance: identity, access, guardrails, human approvals, cost governance and audit trails

Thursday, September 24, 2026

cost saving The harness is worth more than the model: tools add 20+ points to every model tested

cost saving Retender bulk transcription and voice-agent spend while prices reset

cost saving Re-benchmark to the cheapest model that still clears your own quality bar

risk to avoid Price permitting delay into any AI data-centre underwriting

risk to avoid Never sign a flat-fee or per-seat agent contract priced off token math

risk to avoid Mature-node supply keeps growing on domestic Chinese equity funding

stock Inside Chinese tech, export share has separated winners from losers this year

small business A visible ten-year order book: AR5 production scaling from six aircraft to 24 by 2029

Tuesday, September 22, 2026

stock Training-data labelling still compounding: Snorkel AI raised $350M at $3.5B; the company claims 18x growth to a $375M run rate

small business Securing AI agents (permissions, tokens, audit) after a Mac zero-day in a mainstream agent

risk to avoid Platforms are blocking third-party shopping agents; agent products that assume access carry a business risk

cost saving Open-weight models at ~9/10 of frontier capability for a tenth of the price

stock Meta up ~25% on the month on Muse adoption, before the Amazon block and the zero-day

small business Independent evaluation and red-teaming demand: California wants auditors inside labs; Anthropic and Accenture pledged $1B over five years

stock Anthropic IPO moved to November at a reported ~$2T ask