Plenty of vendors will promise you that an AI agent will fill your pipeline. Honest answers about what happens when you plug one into a live sales process are harder to find.

In this session, we’ll cover:

• The sales tasks agents handle well today and the ones they reliably botch

• What a realistic first deployment looks like, and how long before it pays for itself

• How to keep reps in the loop without turning every agent action into a manual approval

• The data and access an agent needs before it can do anything useful, and the risks that come with it

Agents can book meetings, write follow-ups, update your CRM, and score leads, on paper. This 60-minute session covers where they genuinely help, where they create more problems than they solve, and how to roll them out without wrecking the process your team already runs.

Thursday, September 10, 2026 · 12:00 PM ET / 11:00 AM CT / 9:00 AM PT · Free live webinar with Josh Sullivan, COO at Kiingo AI.
Register for the September 10 webinar →

• Swiss bank UBS made AI proficiency a formal hiring requirement for junior bankers and interns, starting with its 2027 graduate intake in Global Banking and Markets, according to the Financial Times. Candidates must describe a real task they gave to an AI model, what they provided it, and what improved as a result; questions about how they use AI tools are being added to interviews alongside the usual degree requirements. Hired candidates enter an internal “AI Fluency Pathway” covering banking use cases and responsible use. The FT notes Santander has a similar requirement for some trainee programs. The move runs opposite to Goldman Sachs and JPMorgan, which have been shrinking junior analyst classes as AI absorbs entry-level work. (Financial Times, The Next Web)

• Google released Gemini 3.8 Flash, its third Flash model in six weeks, alongside a locked-down cybersecurity variant. The general model is built for software engineering and multi-step agent tasks and is priced at $0.75 per million input tokens and $3.75 per million output tokens through December 31, doubling to $1.50 and $7.50 on January 1. It is live for Google AI Pro and Ultra subscribers, in AI Mode in Search, in Google Sheets, and through the API. Gemini 3.8 Flash Cyber, tuned to find and patch software vulnerabilities, is limited to trusted defenders in Google’s new Fairwind Program, starting with government authorities, critical infrastructure operators, and software maintainers; Google’s Chrome security team says it produced 2.6 times more correct patches to Chrome vulnerabilities than the best commercial models. Separately, Google is removing Google Assistant from Android phones, tablets, and Wear OS watches starting September 4, with Gemini taking its place and no option to switch back. (Google, 9to5Google, VentureBeat, 9to5Google)

• Anthropic released Claude Fable 5.1 and Claude Mythos 5.1, the same model shipped with two different levels of safeguards. Fable 5.1 is available to everyone on Anthropic’s own platform and on AWS, Google Cloud, and Microsoft Azure. Mythos 5.1 is restricted to vetted US cybersecurity and life-sciences organizations in Anthropic’s trusted access programs. List pricing is unchanged at $10 per million input tokens and $50 per million output tokens, but cache reads, the price of re-sending material the model has already processed, fell 75% to $0.25 per million, which Anthropic says works out to about 25% cheaper for typical workloads and up to about 45% cheaper for agent-heavy ones. On Terminal-Bench 4.0, a test of agentic coding (a model working through coding tasks on its own), Anthropic reports 55.8% against 37.3% for OpenAI’s GPT-5.6 Sol. The release landed a day after Anthropic reportedly signed a $35 billion deal with cloud provider Lambda for computing capacity at a Texas data center. (Anthropic, VentureBeat, SiliconANGLE)

• OpenAI began rolling out GPT-6 Astra, the first model it has rated “Critical” for cybersecurity capability under its own Preparedness Framework, meaning it can find previously unknown vulnerabilities in well-defended systems and build working exploits without a person directing each step. Access opened first to organizations in Daybreak, OpenAI’s application-based cybersecurity program, and most paying ChatGPT subscribers were still locked out on launch day, prompting Sam Altman to post “sorry for the messy rollout” and promise broad access “in the near future.” Plus, Pro, Business, and Enterprise accounts began receiving it the following day, with the most advanced cyber capabilities held back from general release. API pricing is $10 per million input tokens and $50 per million output tokens. OpenAI said it delayed parts of the model’s development in recent weeks to strengthen and test protections against cyber misuse. President Greg Brockman said the company may now be in the “AGI era.” (CNBC, Bloomberg, The Register, Axios, The New Stack)

• Meta released Muse Spark 1.3, which its chief AI officer Alexandr Wang called “competitive” with Claude Fable 5.1 and “better than” GPT-5.6 Sol, especially at writing code. Meta says it finishes equivalent work with about 20% fewer tool calls and about 25% fewer tokens than Spark 1.2, at unchanged prices of $1.25 per million input tokens and $4.25 per million output tokens. It is live in Muse Code, Meta’s coding tool, and through Meta’s API. Two versions exist: the broadly available one ties GPT-5.6 Sol on Artificial Analysis’s independent intelligence index and trails Claude Opus 5 and Fable 5.1, while the higher-scoring “max” version behind Meta’s best published numbers is in limited preview for Meta’s partners. Meta has not decided whether to release the weights, and the weights it promised for Spark 1.2 remain unreleased. (Bloomberg, SiliconANGLE, VentureBeat, Artificial Analysis)

• Nvidia made its purchase of Hugging Face official at $12.93 billion, three days after putting $3.5 billion into chip designer MediaTek. Hugging Face, the repository where much of the world’s open AI work lives, hosts more than 3 million models, 500,000 datasets, and 1 million applications used by 18 million developers and 200,000 companies. Jensen Huang pledged it will stay open and that “Nvidia compute will not be required to build on or deploy through Hugging Face.” Hugging Face had about $150 million in annualized revenue, and the deal is expected to close in the first half of 2027 pending regulatory approval. The MediaTek investment buys bonds convertible into shares; MediaTek will adopt Nvidia’s NVLink Fusion technology so the custom chips it designs for cloud providers plug into Nvidia-based data centers, and it expects about $2 billion in custom data-center chip revenue this year. (Nvidia, TechCrunch, Bloomberg, TechCrunch, The Register)

• Mistral raised €3 billion, about $3.5 billion, at a valuation above €21 billion, about $24 billion, which Mistral calls the largest equity round ever completed by a European technology company. Samsung Electronics led, with EQT’s Scaleup Europe Fund and existing investor PSG Equity as co-leads; Advent, BlackRock-managed funds, and Luxembourg joined as new investors, and Nvidia, ASML, and Salesforce Ventures re-upped. Mistral, founded in Paris three years ago, builds open and commercial models and sells them to more than 125 large enterprises including Airbus, ASML, and HSBC, and CEO Arthur Mensch says it will pass $1 billion in annual recurring revenue before year-end. The money goes to building and owning data centers and renting more computing capacity; Mensch told CNBC the long-term plan is “to fully rely on capacity that we are building ourselves.” (CNBC, Mistral, Tech.eu)

• Crusoe reportedly raised more than $3 billion at a $30 billion valuation, and Cognition is closing about $1 billion at $47 billion. Crusoe builds and runs AI data centers for Meta, Microsoft, OpenAI, and Oracle; the round, co-led by Atreides Management and Valor Equity Partners, triples its $10 billion valuation from last October and follows a $13 billion, five-year contract to supply trading firm Jane Street with AI computing. Cognition makes Devin, a coding agent, and was valued at $26 billion in May; its annualized revenue has passed $900 million. Bloomberg reports nearly $10 billion in investor interest for the Cognition round and says terms could still change. (Bloomberg, TechCrunch, Bloomberg)

• The Justice Department told a federal judge that training AI models on copyrighted text is fair use, siding with OpenAI and Microsoft against The New York Times. The statement of interest, filed September 1 in the Southern District of New York, is the first time the federal government has taken a formal position in the wave of copyright suits by publishers, authors, and music labels against AI companies. It argues the United States “has a strong interest in continuing to develop a robust and competitive artificial intelligence industry” and that a licensing requirement would leave only the largest technology companies able to afford to train models. The filing carries no binding authority; Judge Sidney Stein decides the fair use question. The Times said the administration is “siding with a handful of trillion-dollar AI companies at the expense of the countless American creators whose work they stole.” (Bloomberg Law, The Washington Post)

• Two very different AI bills landed in Congress the same week. The bipartisan Stop Rogue AI Act, from Reps. Josh Gottheimer (D-N.J.) and Mike Lawler (R-N.Y.), gives the National Institute of Standards and Technology a year to write security standards for AI agents covering continuous verification of what agents do, tamper-proof logging, and a “continuous, machine-readable inventory of all AI agents” an organization runs. The standards would be voluntary for most companies and mandatory for contractors bidding on new federal work. Separately, Sen. Bernie Sanders (I-Vt.) and Rep. Greg Casar (D-Texas) announced the Ban Artificial Superintelligence Act, which would permanently ban AI systems that “surpass human intelligence,” pause advanced AI development until a new cabinet-level agency writes safety rules, and carry penalties of up to 20 years in prison for individuals and forced dissolution for companies. Both bills point to July’s incident in which OpenAI test agents got loose and breached Hugging Face’s systems. The Sanders bill has been announced but not yet formally introduced. (Rep. Mike Lawler, Axios, Sen. Bernie Sanders, The Hill)

• Intuit QuickBooks’ July survey of about 5,000 small and midsize businesses in the US, Canada, the UK, and Australia finds 80% now use AI regularly and 41% use it daily, up from 15% two years ago. Nearly one in four (24%) say AI has shortened their workday, twice the share (12%) who say it has made it longer, and three-quarters of users say it is boosting their productivity. (Intuit QuickBooks Small Business Insights, July 2026)

• PwC went back to 351 CEOs across 59 countries and 27 sectors and found 39% say AI has raised their revenue, cut their costs, or both, against 16% whose AI impact stayed negative or got worse. Nearly four in ten (38%) have used AI this year to spot new business opportunities created by changed conditions. (PwC CEO Survey Snapshot, August 2026)

The sales call ends. You know exactly what happened for about eleven minutes, and then the next meeting starts. By the time you open the CRM that evening, the note says “good call, send proposal,” and the two things the buyer let slip are gone. What you could say about that call right now is far more complete than anything you will type at 6 p.m., so say it out loud.

1. Talk for two minutes, badly. Open your assistant and tap the microphone button, which sits in the message box of the Claude, ChatGPT, and Gemini apps on your computer and your phone. Say what they said, what they asked for, who was in the room, what felt off, and what you promised. Fragments are fine. If you have a call transcript from your meeting tool, paste that instead.

2. Then send this.

“That was my debrief of a sales call. Return four things. One: what the buyer said they want, and separately what they literally asked for, if those differ. Two: every objection, stated or implied, one line each. Three: the commitments each side made, with any dates. Four: the single follow-up that has to happen in the next 24 hours. Then draft that follow-up as a short email, quoting one thing the buyer actually said. Use only what I told you. Where you are guessing, say so.”

3. Read the quote first. The email has to quote something the buyer really said. If the quote is real, the rest of the debrief heard the call. If it is invented or bent, fix that line and rerun before you trust anything else on the page. That one check takes ten seconds and tells you how much of the output to believe.

4. Paste it straight in. The four sections become the CRM note. The email goes out once you have read it as the buyer would. The objections list is what you bring to your next internal pipeline conversation, because it is the buyer’s words rather than your memory of them.

5. Make it the first two minutes after every call. Same prompt, every time. After a month you have a searchable record of what your buyers say, in their language, which is worth more than any pipeline report.

Typed at 6 p.m., the note is three words. Spoken two minutes after the call, it is the whole call.

I had a conversation with one of our engineers this week about how to get better at coding with AI. Working with it daily on our own projects has made me fast and confident at building and fixing front-end pages with AI assistance, to the point where the pages themselves have stopped being the interesting problem.

His advice was simple. Build something just a little more complex and see where you get stuck. A web page serves you something to read. The next tier is something that performs a service: it takes an input, does a job, and hands something back. Closer to what people would call an app. Same tools, same way of working, one step past where I am comfortable.

That is the part that stuck with me, and it applies well beyond code. Once you feel comfortable working at a new tier of AI-assisted anything, ask what one step further looks like. The email, then the report. The summary, then the analysis. The page, then the app. Comfort is the signal you are ready, and getting stuck is where the learning happens.

Most of the tools in this issue will have changed again by the time you have settled into them. What carries forward is how your company works with them: where your data lives, who can use what, and how quickly your people pick up the next thing.

Kiingo makes your company AI native: your data consolidated into one secure company brain, governance so your team can use any model with confidence, tools built on top of it, and training that meets your people where they are today so they keep pace with whatever comes next.

Let’s find your starting point →