Checking, Not Guessing

Hello,

This is Simon with the latest edition of The Weekly. In these updates, I share key AI related stories from this week's news, list upcoming events, and share any longer form articles posted on the website.

I use Claude frequently during my working day for a varied number of tasks. It might be to ask a question about a specific feature in our product, to ask it to highlight my outstanding actions from my inbox, or to help build a presentation deck. Regardless of the actual task, it's fair to say that Claude is now firmly a part of my working day.

Even when I'm on a call with a client, it's incredibly helpful to ask Claude a question and have it running in the background while we're speaking. That way, I can give a decent answer on the call itself, rather than having to end the call, spend time researching the answer, and then follow up a few hours (or even a day) later.

What makes Claude so useful for me is having it connected to most of our other platforms and systems such as email, Slack, Jira and Gong. This gives me confidence that it's checking, not just guessing, when it gives me an answer. Without those systems connected, Claude would be answering based purely on the knowledge within the model itself, or whatever it can currently find on the internet — and as we all know, that can't always be trusted.

Your day-to-day role is likely different to mine, but you should be able to find a cut-off point where you stop relying on AI and turn to a human expert instead. Even with Claude providing quality answers I can use with clients, I have a limit to how far I'll take it. If the topic is truly technical, I'll draw the line and refer to one of my technical colleagues for the best answer. If a client has an issue with the platform, I'll direct them to our Support Engineers.

Right now, I feel I've got the balance right: using Claude for quick, easy answers I could probably find myself but that would take longer, and turning to human experts when it's a topic I can't answer myself — and therefore can't judge whether the AI's response is accurate.

Leadership is asking: are we getting value from AI? Which tools are worth the spend? Where are we exposed? Right now, most teams have no idea.

Harmonic Security Usage Explorer changes that.

You get a complete picture of how your organization uses AI, automatically categorized into custom tasks and use cases.

You’ll see the projects being worked on, who’s using what tools, where AI investments are driving value, and where employees are engaging in risky behavior.

CIOs can rationalize spending and cut wasted licenses. CISOs can pinpoint where risk exists and neutralize it. AI committees can show exactly how their efforts are paying off.

See it in action

OpenAI's CFO Sarah Friar told investors on 14 August that enterprise revenue has overtaken consumer subscriptions for the first time, reaching an annualised pace of $40 billion — two full quarters ahead of the company's own forecast. Business customer revenue rose 32% in a single month, according to reporting by multiple outlets including TechTimes. OpenAI now counts more than 2 million business users paying via API or enterprise licences.

Why it matters: The shift signals that the commercial centre of gravity in AI has moved from consumer novelty to business infrastructure. If your organisation hasn't formalised its AI vendor relationships, you're now operating in a market where enterprise deals are setting the pace and the pricing.

Anthropic told investors last week that its annualised revenue run rate has surpassed $65 billion with Q2 quarterly revenue of $11.5 billion representing a 14-fold increase on the same period last year, according to Bloomberg (17 August). The company, which makes the Claude family of models used heavily in enterprise settings, filed confidentially for a US IPO in June and is expected to list as early as October. Investor valuation expectations now exceed $2 trillion.

Why it matters: Anthropic is no longer a research lab with enterprise ambitions — it's one of the largest software businesses on the planet, measured by revenue growth. For businesses using Claude in production, this trajectory matters: it signals financial durability, but also impending public-company pricing discipline and shareholder pressure to monetise enterprise relationships more aggressively.

Gartner's research, widely cited in enterprise and investment circles throughout July and August, predicts that more than 40% of agentic AI projects will be cancelled before reaching production by the end of 2027. The reasons are not technical: projects stall in integration, consume budget without producing output, and get killed when governance frameworks can't keep up. Only 23% of organisations report meaningful ROI from AI agents — lower than from standard generative AI. A separate survey found that just 21% of companies have a mature governance model for autonomous AI.

Why it matters: Agentic AI is where most enterprise AI investment is heading. The failure rate is a signal that the gap between demo and production is much wider than vendors suggest. Before signing off on the next AI agent project, it's worth asking not just "can it do the task?" but "who owns it when it goes wrong?"

Thanks for reading, and see you next Thursday.

Simon,

Was this email shared with you? If so subscribe here to get your own edition every Thursday.

Enjoying Plain AI? Share it and get a free gift!

If you find this newsletter useful, why not share it with a couple of friends or colleagues who would also benefit? As a special thank you, when two people subscribe using your unique referral link, I’ll send you my "Exclusive Guide: Supercharge Your Work with NotebookLM." It’s a practical, no-nonsense guide to help you turn information overload into your secret weapon.

You can also follow me on social media: