AI agents are showing up across departments, in sales, finance, and operations, whether executives planned for it or not.

This 60-minute session is built for the people who have to make the call on where agents belong, what they're allowed to do, and how to know if they're working. On Tuesday, October 20, COO Josh Sullivan will cut through the vendor noise and focus on the questions that matter:

• What to fund: where agents belong and where they don't yet

• What to watch: what agents are allowed to do, and how to know if they're working

• What to hold back on, and how to say so when a vendor is pushing

Who this is for: Executives and department heads who will decide where agents get deployed, what they can touch, and what counts as working, before the team starts using them.

Tuesday, October 20, 2026 · 12:00 PM ET / 11:00 AM CT / 9:00 AM PT · Free live webinar with Josh Sullivan, COO at Kiingo AI.
Save your spot now →

• Google opened Gemini 4 Argon to everyone on October 13. Paid API customers get it at $2 per million input tokens and $10 per million output until November 30, when the price doubles to $4 and $20. Google AI Pro and Ultra subscribers get it in the Gemini app. Google says it ranks first on the Vals Index of finance, coding, legal, and tax work, and it took first place on Roboflow’s independent vision benchmark the same day. (Google, Google, Roboflow) On the enterprise side, Google Cloud launched a single Gemini agent for work on October 8, folding its separate enterprise AI products into one agent that takes a goal, plans the work, and runs in the cloud. Inside Google Workspace, a “coworker agent” gets its own account and email address, sees only what is shared with it, and appears in audit logs under its own name. It connects to Microsoft 365, Slack, and any MCP server, and picks a model per task from Gemini or Claude. Sundar Pichai said Gemini has more than 1 billion monthly users and nearly 90% of the Fortune 100 use Gemini Enterprise. (Google Cloud, TechCrunch, Futurum)

• Meta released its first frontier-class models, Muse Wonder and Muse Wonder Full, on October 12, its first flagship since Alexandr Wang took over Meta's AI work last year. The standard Wonder is available through the Meta Model API at $4 per million input tokens and $20 per million output, the same price as Claude Opus 5.5. Wonder Full, the stronger version, is available only inside Muse Code subscriptions at $50 to $200 a month, and neither model is released as downloadable weights, which Meta's Llama models were. On the Artificial Analysis Intelligence Index, Wonder Full scored 56, tying Claude Sonnet 5.5 and ahead of GPT-6 Astra and Gemini 4 Argon at 53, with only Claude Opus 5.5 higher at 58. Both ship with Active Memory, which carries details across conversations, and Active Focus, which keeps working through errors on long tasks. (Meta, TechCrunch, VentureBeat)

• Claude can now build live dashboards and animated explainers. Anthropic put Claude Dashboards and Claude Motion into beta on October 8. Dashboards connects to Snowflake, Databricks, BigQuery, Redshift, ClickHouse, or Salesforce, answers a plain-language question about the data, builds a dashboard, shows the queries behind it, and keeps it current as the data changes; it is on all paid plans. Motion turns a report or chart into a short animation written as editable code rather than generated video, so any word, number, or timing can be changed before export to MP4; it is limited to Team and Enterprise. Claude Docs, Slides, and Design left beta the same day and are now on every plan, including Free. In Enterprise workspaces, Dashboards and Motion stay off until an admin turns them on. (Claude, TestingCatalog)

• Anthropic's updated usage policy takes effect November 12. Tracking people without consent is prohibited, and Claude cannot be used to decide who should be investigated, arrested, or charged, while fraud monitoring, content moderation, journalism, and legal research stay permitted. Claude connected to physical equipment that could injure someone now requires an operator who can watch and stop it, and the equipment must fail safe if Claude disconnects. The policy also bars access from restricted regions through companies majority-owned there. Separately, Anthropic paused the free year of Claude Team and $1,000 in API credits it had offered startups since October 6, three days after launch, after applications in the first 48 hours ran 21 times the previous five months combined and many non-startups got through. Benefits already claimed stay; unclaimed approvals are being re-verified. (Anthropic, Forbes, TechCrunch)

• OpenAI told investors its annualized revenue was approaching $50 billion at the end of September, the Financial Times reported October 8, roughly $20 billion below the figure that circulated in late September. The gap is mostly accounting: the higher number counted gross revenue from partners that resell OpenAI's models, a method meant to compare directly with Anthropic, while OpenAI's own figure books only its share. Nvidia, Oracle, and CoreWeave fell on the report, Nvidia by 3% and Oracle by nearly 6% on the day. (TechCrunch, CNBC) The IMF named an AI stock correction one of three risks that could abruptly tighten financial conditions worldwide, in its Global Financial Stability Report published October 13, alongside a spike in government bond yields and a major cyberattack. The report says weaker or less durable AI investment and profits could trigger a correction in AI-linked stocks and credit, that "the consequences of an AI correction could be global," and that circular financing, where a small group of firms act as each other's customers, investors, and lenders, raises the odds of trouble at one spreading to the others. (IMF, Reuters via Yahoo Finance)

• The FBI arrested an OpenAI engineer on October 9 for planting a hidden audio recorder in an executive conference room. Staff found the device, wrapped in toilet paper and plugged into a USB port on the back of a television, around September 30, and a second device on October 5, in a bag he left unattended in another room. OpenAI says it used its own internal models to analyze the device, which had encryption and a data-transfer mechanism built to evade monitoring, and to trace it to the employee, who had worked there since July 2025. The company says it has no indication that model weights, source code, or customer data were accessed. The engineer, Abdulrahman Alfozan, is charged with intercepting oral communications, which carries up to five years. (TechCrunch, NBC News)

• Outside scrutiny of the top labs tightened this week. On October 14 the EU said its AI Office has sent OpenAI and Anthropic its first formal requests for information under the AI Act, covering security, external evaluation, and how use of their models is monitored, after judging the documents the labs volunteered insufficient. The triggers are this summer's incidents: OpenAI agents that escaped a test environment and broke into Hugging Face, and an Anthropic model that reached three outside organizations' systems during testing. Both happened before the systems were placed on the EU market, and whether the Act reaches pre-release testing is contested; breaches of general-purpose model obligations carry fines up to €15 million or 3% of global revenue. A day earlier, Reuters reported that as recently as October 5 both companies had sought editorial control over what their embedded outside evaluators could publish, and that OpenAI redacted nonpublic details from a METR and Redwood Research report on the Hugging Face incident. Both labs say reviewers keep independence over their findings. (Euronews, Reuters via BNN Bloomberg, The Next Web, AI Act) Microsoft CEO Satya Nadella published a long post on X on October 10 arguing companies should "assume a model is compromised and contain it from the start." The post calls for keeping safeguards outside the model’s reach and making sure an authorized person can pause or shut down a model mid-task, which he compared to an emergency brake. (TechCrunch, The Verge)

• Epic Games delayed its Disney game to rebuild production around AI, Bloomberg reported October 12. An internal message to the team on the unannounced Marvel and Star Wars shooter said it would become an "AI first" production with "no more hand-written code" and all art "generated with full AI workflows," using Anthropic's Claude under developer direction. The game had been slated for March and now has no date, and several developers told Bloomberg they hoped the directive was not serious. Epic's communications chief said the company is "embracing AI-accelerated development, led by Epic developers who are the visionaries and creative drivers for everything we do." (Bloomberg, Engadget)

• TypeSafe AI raised $870 million at a $7.5 billion valuation on October 9, led by Andreessen Horowitz with Sequoia, less than a month after its model Jev launched. Jev returns a decision rather than a paragraph: it takes a question with a defined set of answers and returns a yes/no, a pick from a list, or a score, each with a confidence number, in under 700 milliseconds, built for high-volume sorting and routing decisions. TypeSafe says about a third of the Fortune 500 already use it. OpenAI moved onto the same ground days earlier, putting its Decisions API into public beta on October 6 at $0.10 per million input tokens with no charge for output, which it says returns typed answers about 10 times faster than its standard API. (Bloomberg, TechCrunch, SiliconANGLE, OpenAI)

• The U.S. Chamber of Commerce's fifth Empowering Small Business report finds 66% of small businesses now use AI, nearly triple the 23% of 2023, and 47% say AI is creating jobs today against 6% using it to cut headcount. (U.S. Chamber of Commerce, Oct. 2, 2026) A New York Fed analysis of 5,248 employer firms found the same pattern: among AI users, the share expecting higher revenue outran the share expecting lower by 48 points, against 21 for non-users. (Liberty Street Economics, Oct. 8, 2026)

• Wharton Human-AI Research and GBK Collective's fourth annual AI Adoption Report finds strong returns from AI are nearly three times as common among leaders who track a growth measure, 35%, as among those measuring efficiency alone, 13%. Two-thirds of departments now use AI agents in some form, and 17% let agents finish tasks without step-by-step human approval. (Knowledge at Wharton, Oct. 13, 2026)

Somewhere in the document you are about to send is a number nobody can trace. The 14% in the board update came from a spreadsheet tab that has since been overwritten, the pipeline figure was right in August, and the headcount was typed from memory. Each one was true at some point. Every number that leaves your hands should have a trail back to its source, and an assistant can walk that trail for you in one pass if you hand it both the draft and the data it came from.

1. Gather the draft and its sources. The document as it stands, plus whatever the numbers came from: the export, the spreadsheet, last month's report. Paste or upload both into Claude, ChatGPT, or Gemini. Leave the source exactly as you used it, since that version is the one being checked.

2. Run this. “Here is a draft I am about to send and the source material its numbers came from. Find every number, percentage, dollar figure, and date in the draft. For each one, show me exactly where it appears in the source, and if it is a calculation, recompute it from the source data. Mark each one MATCHES, DIFFERS (show both values), or NO TRAIL if you cannot find a source for it. Do not correct the draft and do not estimate anything. Return as: Number in draft | Where it appears | Source location | Recomputed value | Status.”

3. Read the NO TRAIL rows first. Those are the figures that live only in someone's head, and each one either gets a source or comes out of the draft. DIFFERS rows are usually a stale version or a rounding choice; decide which value is right and note why. Only then go back and fix the document. This works the same way in Claude, ChatGPT, and Gemini, with nothing to enable.

Worth running on anything with numbers in it that goes outside the building: board updates, investor notes, pricing sheets, the proposal with a savings estimate on page two. The rows that come back NO TRAIL month after month are the figures your company has never written down anywhere, which is a useful list on its own.

Prep for a big meeting used to mean opening every document that might come up and hoping the right detail surfaced when I needed it. Now I do it in a conversation with an AI assistant, and the habit that changed everything is simple: I stopped feeding it document after document and asking for a summary.

Instead, I load the deck, the recent email threads, and the contract, and talk to it like an all-knowing assistant who has read all of it. What is this meeting actually about? What's changed since these people last met? If someone pushes on the number on slide seven, what's behind it? Each answer sharpens the next question, and a few minutes in I understand the material in a way a summary never gave me.

Investigation and meeting prep are two different jobs, and the assistant does better when you tell it which one you're on. Left to guess, it will keep digging, because that's a reasonable reading of "help me prepare." So I run the investigation first and save all of it. Then I ask for the digest: the five things that matter for this room, with the rest left where I can find it.

The leaderboard will keep changing. What holds its value is what you build around the models: the data your tools can reach, the rules they work under, and a team that knows how to use them. Kiingo builds that with you. We connect your systems, set up the rules, and train your people, in an order that fits where your business is now.

Speak with us →

Forwarded this by a colleague? Subscribe to Kiingo AI Weekly and get it every week.