Hello fellow keepers of numbers,

OpenAI launches GPT-5.2 because they can’t stand not being the center of attention. Which will really piss them off when they find out Accenture partnered with Anthropic just one week after their partnership with OpenAI. Also, KPMG becomes the first major accounting firm to receive ISO certification for AI governance.

Plus, stick around for a cool example of how to use Claude Code to work with files on your computer (no coding involved, I promise).

THE LATEST

OpenAI introduced GPT-5.2, a performance-focused update to the GPT-5 family aimed at faster responses, lower latency, and more reliable reasoning. The release is positioned as an upgrade for day-to-day professional work like document analysis, coding, and longer, multi-step tasks rather than a flashy new feature drop.

OpenAI is emphasizing improvements in practical productivity tasks. The company highlights stronger performance in generating and analyzing spreadsheets and slide decks, better code generation, and improved understanding of visual content like dashboards, PDFs, and tables. Benchmarks show GPT-5.2 Thinking outperforming GPT-5.1 on math and abstract reasoning tests, while producing factual errors roughly 30 percent less often on internal evaluations.

For developers, GPT-5.2 is also available via the API at higher prices than GPT-5.1. OpenAI’s system card states that GPT-5.2 is priced at $1.75 per million input tokens and $14 per million output tokens, with a 90 percent discount on cached inputs. In ChatGPT, the new model is rolling out first to paid subscribers, while earlier GPT-5-series models like GPT-5.1 remain available in the near term through the API and as legacy options for some users.

Why it’s important for us:

This is the immediate result of the “code red” issued by CEO Sam Altman. It’s also more proof that OpenAI and other AI providers are sitting on better models waiting to release them publicly.

As I’ve said for all other model releases, I put no weight on the model benchmarks. Most accountants aren’t pushing the models to the limit, so the response style and tone of the model mean more to me than benchmarks when I evaluate it.

TBD on this model update until I’ve had more time to use it. But I was pleasantly surprised with GPT-5.1. If it’s similar to GPT-5.1, but slightly smarter, I think I’ll probably enjoy GPT-5.2 as well.

Anthropic and Accenture announced a multi-year expansion of their partnership, creating a new Accenture Anthropic Business Group focused on moving enterprises from AI pilots into full production. About 30,000 Accenture professionals will be trained on Claude, making this one of the largest groups of Claude practitioners globally. The group will specialize in deploying Claude models inside client environments, with a focus on regulated industries like financial services, life sciences, healthcare, and the public sector.

Accenture will become a premier AI partner for Claude Code, Anthropic’s coding assistant that helps developers write, refactor, and understand software. Claude Code will be rolled out to tens of thousands of Accenture developers as part of a new joint offering for CIOs that puts AI at the center of the software development lifecycle, complete with productivity measurements, workflow redesign, and change management support.

The companies are also co-developing packaged industry solutions for compliance-heavy use cases, such as automating document-heavy regulatory workflows in banking or modernizing legacy systems in public sector agencies.

Why it’s important for us:

Well, literally just last week Accenture partnered with OpenAI. Now they’ve announced a partnership with Claude. This is a great move by Accenture. There are a lot of good options available when it comes to companies deploying AI models and AI agents. Accenture has made sure they’re experts and preferred partners for the two largest providers in the market.

Beyond the fact that this is interesting because it comes literally one week after the news of the OpenAI deal, it’s also interesting how they’re planning to deploy AI models and agents. They’ve created a partnered business group where it sounds like they’ll brainstorm and explore how to deploy AI models, and more interestingly, AI agents across large companies.

Seems like we’re often hearing about how AI models and agents fail during implementation these days. It makes a lot of sense to me for firms like Accenture to employ experts capable of helping firms with these large-scale implementations.

The promise of AI agents is great, and they can even do very cool things right now with current capabilities, but it’s difficult for most firms to successfully implement them without the support from an expert. And even if they’re implemented, evaluating the success and ROI can often be difficult.

If you take nothing else away from this news… Accounting firms and the accounting industry are sleeping on Claude.

KPMG in the U.S. announced it has earned ISO/IEC 42001 certification for its organization-wide AI management system after an independent audit. The certification validates KPMG’s processes for AI governance, risk assessment, internal controls, and continuous improvement across how it develops and uses AI. KPMG says this makes it the first major professional services firm in the U.S. to achieve this AI-specific management system certification.

ISO/IEC 42001 is the first international standard for an Artificial Intelligence Management System (AIMS), setting requirements for how organizations establish, implement, maintain, and continually improve governance over AI systems rather than evaluating individual models or tools. The standard defines an AI management system as the policies, objectives, and processes an organization uses to ensure responsible development and use of AI, including how it manages AI-related risks and opportunities across the business. It is designed to sit alongside other ISO frameworks like ISO/IEC 27001 for information security.

The KPMG certification covers governance and management systems, not model accuracy or performance. In practice, that means auditors reviewed how KPMG inventories AI systems, assesses AI risks, applies controls like data protection and bias checks, and monitors AI over time. The certificate is valid for three years and is subject to ongoing surveillance audits under the ISO framework, which require KPMG to show evidence of continuous improvement in its AI governance processes.

Why it’s important for us:

This felt like an important story to cover for two conflicting reasons:

1) AI governance is really important

This is stating the obvious, but firms need to work on their AI policies ASAP, and they need to spend time thinking about AI governance. Staff are likely using AI whether the firm has rolled out official business or enterprise AI licenses or not. Beyond that, AI is now being used within most of the major existing software tools in the industry.

Good for KPMG for being ahead of the industry in this regard.

2) What does this certification really mean?

To completely and immediately contradict myself and show my cynicism, does a framework like this really matter right now? Maybe for a large firm like KPMG, it does. But if you’re delaying major AI rollouts or initiatives while you wait months or years for your AI governance to get certified, then you’ve already lost.

And here’s what this certification does NOT mean:

It’s just a framework to assess the risk and how to properly use AI. But AI is also moving so incredibly fast that I don’t know how any framework could possibly last 3 years, let alone 3 months right now.

I’m not saying this certification is useless. Good for KPMG for leading the way in frameworks for responsible AI. But also, this type of bureaucratic red tape is why so many accountants (like myself) have chosen, and are still choosing, to leave the Big 4. [End of cynicism, for now]

PUT IT TO WORK

Disclaimer: Coding knowledge not required

I feel like I have to make that disclaimer any time I’m about to mention Claude Code.

I want to share a use case for Claude Code that does not include coding custom apps, vibe coding dashboards, etc. I think some people hear Claude Code, Cursor, or other IDEs and immediately think that must be for software engineers. I’m here to prove that wrong.

The Loom I’ve shared below explains how you can link a folder on your computer to chat with your files and create documents. If your first thought is “Well, I can just attach files in my conversations in Claude, ChatGPT, or Copilot”, I address that in the video. I think you might be impressed with the quality difference between the two.

Using Claude Code to chat with and create documents - Watch Video

WEEKLY RANDOM

Attn Google Workspace users: Google recently released Google Workspace Studio, which is their internal automation software. The UI looks fairly similar to Zapier, but it’s built mainly for the GSuite tools. It also looks like it offers a few native connectors.

If you already use Gems (same thing as ChatGPT projects), the automations can actually call existing Gems to complete tasks. Below is one of the examples they provided to analyze customer service requests and draft a response.

It seems like there could be quite a few quick wins built around Gmail. For anyone looking to clean up their inbox, spend less time on emails, or trigger other workflows based on certain emails received from clients, this could be a useful tool. Google has also provided some base templates to use as starting points.

Combining Gems with automations built in Google Workspace Studio could be a very powerful tool. This is an extremely interesting new product. It’s not currently taking the place of your third-party automation software (n8n, Make, Zapier, etc.), but it might be a great tool to use for any GSuite-specific automations.