Hello fellow keepers of numbers,

The agents just…never…stop. Everyone told us this was the year of AI agents, so I guess we should’ve known. The problem is most of them suck. On an unrelated (maybe?) note, Intuit launched their Intuit Accountant Suite, which introduces AI and AI agents throughout the application. BILL also announced two new AI agents to help with W-9 collection and receipt reconciliations. Lastly, Digits opened their API to allow for custom integrations.

THE LATEST

Intuit announced the retirement of QuickBooks Online Accountant (QBOA) and the launch of Intuit Accountant Suite at its recent Intuit Connect conference in Las Vegas. The new suite includes features like Books Review (close lists), KPI dashboards, anomaly detection, and capacity/task management.

Intuit Accountant Suite is now generally available to US-based accounting firms. Premium features are available in a beta program at no cost, with fees expected to begin in early 2026. QBOA will sunset at the end of calendar year 2026. The suite bundles ProAdvisor Academy content and provides tighter in-app workflows designed for firm teams.

Intuit describes the new platform as an AI-powered operating system, moving beyond QBOA's traditional role as a client portal. The suite is designed to manage practice staff, workflows, and profitability while providing consolidated client management and workflow automation features. Books Close beta is currently available at no extra cost, but pricing will be introduced when the feature becomes generally available.

Why it’s important for us:

In one of the least shocking moves of all time, Intuit launched an updated platform intended to be built upon AI tools and features. We’ll continue to see similar announcements by other accounting software providers over the next year.

If the AI features and Intuit Accountant Suite live up to their promises, it’ll be a nice upgrade for a lot of firms right now that struggle to meaningfully use AI or automation in their workflows. In my opinion, there are two key items that stand out from the announcement: (1) AI on top of QBO data and (2) AI agents.

(1) AI on top of QBO data

I think one of the most common questions I get and see others ask is “How do I know the data I put into ChatGPT [or insert AI model of choice here] is secure and confidential?” Obviously a good question.

For those concerned about taking data from QBO and uploading it into an AI model, this might feel a little more secure and natural since it sits within your Intuit account. Don’t be fooled. This data is still going to an AI model, and likely the same ones you would be using anyway (ChatGPT, Claude, Gemini).

You need to develop the same level of comfort with the Intuit AI that you would with any other AI model. The Weekly Random today covers why you should feel comfortable sharing your data with AI models since it’s an important factor to note with this announcement.

(2) AI agents

I would temper expectations around the Intuit AI agents, especially in the early days. Even with today’s powerful AI models, agents often have a much higher failure and inaccuracy rate than a typical human would when performing tasks.

Evaluating the benefits of AI agents is important, especially when agents have the ability to edit your information and/or share data with external parties. The risk is that an AI agent does something incorrect and you’re left fixing the errors. Hopefully Intuit has built-in guardrails for those cases and the ability to integrate human-in-the-loop steps with the AI agents.

The reality is these features are going to be very raw. Several are still in beta, and this only just rolled out to the public. What we’ll need to evaluate in live production:

I’m currently bullish on using software that majors in specific areas and diversifying the tech stack without overspending.

For example, how likely is it that Intuit will be able to build the best AI model for accountants? Very unlikely. That’s not their focus. So if you can use your QBO data and plug that into another software whose business is to build the best AI model for accountants, then you’re likely to achieve better results.

The same goes for other workflows across the firm and tech stack.

Digits announced the launch of Digits Connect API, a free developer interface that gives direct access to their Autonomous General Ledger without partner fees, usage limits, or marketplace restrictions. The API allows developers to send raw transaction data directly to Digits' AI system, which then automatically researches vendors, enriches profiles, and posts fully categorized entries to the ledger. The API is available in preview for developers now, with app submissions opening later this year.

Digits is an AI-native general ledger platform that launched publicly in March 2025. The platform uses AI to automatically categorize transactions, match vendors, and reconcile accounts in real-time.

As part of this launch, Digits also now supports dimensional accounting across departments, locations, and projects. This allows businesses to classify and analyze payroll, expenses, or revenue by different parts of the business without exporting to spreadsheets or managing manual tags. The company describes it as "a simpler, more flexible way to see how money moves through an organization: where it's earned, where it's spent, and how each area contributes to the whole."

Digits positioned the free API as part of their open accounting philosophy: "Unlike legacy accounting platforms, Digits doesn't believe access should come with strings attached. The Digits Connect API is free—no partner fees, no usage paywalls, no marketplace lock-ins."

Why it’s important for us:

The API lets you build custom connections that don’t exist natively. Instead of manually importing transactions, you can build an automation that pulls the data in as transactions are recorded or on a specific schedule.

This is one more step toward an automated close process, which is obviously incredibly valuable. But it also offers better real-time insights for firms and business owners since the automation can be created to pull live data or scheduled daily/weekly.

Digits has been on my accounting software to watch list since their launch. This is a really nice step forward for them, and the industry as a whole.

Most legacy accounting software providers break your kneecaps and steal your wallet just to get access to their API. The unfortunate truth is they don’t really want you using their APIs. They’re poorly built, often break, and allow the company to maintain control over your tech stack. If you buy CCH Axcess and they lock down their API, you’re now more likely to buy other CCH products that provide basic integration features.

New accounting software providers that open their APIs push the industry forward and put pressure on the legacy providers. The Digits API looks very promising. I’ll be interested to hear feedback from users and developers over the next several months.

In addition to the API, Digits also announced dimensional accounting across departments, locations, and projects. This can also be handled by the AI, which is a huge benefit compared to manual or semi-manual tags that a lot of firms use today.

BILL announced the launch of BILL AI on October 28th, a suite of AI agents designed to automate finance operations for small and midsize businesses. The company introduced two primary agents: the W-9 Agent and the Reconciliation Agent.

The W-9 Agent autonomously emails vendors to collect and validate W-9 forms, eliminating over 80% of the manual steps involved in tax compliance. It's the first of its kind in the industry, according to BILL. It is expected to be available in the coming weeks.

The Reconciliation Agent automatically codes transactions and reconciles receipts, providing full visibility into expenses. During early rollout, transactions processed entirely by AI increased 533% with 92% accuracy. It is expected to become available over the next six months.

The agents are trained on BILL's proprietary data from over $1 trillion in transactions and 1.3 billion documents across nearly 500,000 businesses and 9,000 accounting firms.

BILL also announced additional features including agentic-powered onboarding for new users and an agentic-powered BILL Assistant for personalized support.

Why it’s important for us:

These are prime examples of when it makes sense for an existing software to implement an AI feature or AI agent. BILL is targeting specific pain points that their customers have. Unlike other companies that seem to build AI features for the sake of it (*cough cough* Intuit), BILL is spending their time on useful features.

As with any AI agent, the reliability will be a major factor. They’ve noted the W-9 Agent eliminates over 80% of manual steps and the Reconciliation Agent had 92% accuracy. Both results are fantastic, and if the numbers remain the same once fully deployed, it’ll be a huge success. But in the wild, there are often weird edge cases that’ll reduce accuracy and reliability.

The other factor I typically evaluate with built-in AI agents is the ability to add a human in the loop. It’s unclear right now if there’s any customization available in these AI agents.

BILL’s edge is the sheer volume of data available to them to train AI models and agents for these types of tasks. They also seem to be focused on solutions to real problems their customers have. BILL is a very compelling option for firms not already utilizing the software.

PUT IT TO WORK

Steal my template below to use for your own custom projects, custom GPTs, or use in your AI agents.

The template has proven to be rather successful for me across all AI models and AI agents. Other formats for custom instructions can obviously work well, so this isn’t the only approach. But hopefully using this template can yield good results and save you time since you’re not starting from scratch.

You can also provide this template to your AI model of choice and describe what you’re trying to accomplish to have the AI write the instructions for you in this template’s format.

— TEMPLATE —

[1-2 sentences on what the model is an expert at and how it interacts.]

TASK

[What is the job of this AI model / project? If there are multiple tasks, I break them out into steps so the model or agent knows exactly what task to complete and when.]

CONTEXT

[Provide any information the model needs to understand to perform its task well. This could be information about the business, the tasks, the project, etc.]

PROJECT FILES (“TOOLS” if using an AI agent in n8n)

[Describe the information included in the project files so the model understands how and when to use it. I typically provide the exact name of the file or tool, then a sentence or two describing it and when to use it.]

RULES

[Bullet point list of things the model should and should not do. Use short sentences. This can be things like the tone and style of communication, words/phrases it should not use, or other rules that help you tweak the model after testing.]

OUTPUT FORMAT

[Use this section if you need the model to respond in a certain type of output. For example, if you need it to provide you code or JSON, then provide it the format you expect. If you'll just be chatting with the model, you do not need this section.]

EXAMPLE OUTPUT

[If you provided an output format, provide the model with an example here. You can also use this section to provide an example of how you'd like the model to communicate. Typically, I recommend one example, but 2 or 3 can be beneficial if there are a few types of responses you expect.]

WEEKLY RANDOM

One of the most common questions I see today is “How do I know the data I put into ChatGPT [or insert AI model of choice here] is secure and confidential?”

There are a couple of important factors I want to address. I’m going to cover the 5 most common AI providers right now for firms (in my opinion): OpenAI’s ChatGPT, Anthropic’s Claude, Google’s Gemini, Microsoft’s Copilot, and Perplexity.

(1) SOC 2 Type II

OpenAI, Anthropic, Google, Microsoft, and Perplexity are each SOC 2 Type II compliant. It’s important to note that for Google and Microsoft, the report covers a wide array of products, but for both it includes their AI models.

(2) Business and Enterprise plans

The business and enterprise plans for each of the AI providers gives you and your team privacy and confidentiality. OpenAI, Anthropic, Google, Microsoft, and Perplexity do not train on your data on these plans.

(3) Opting out of model training on personal plans

It should also be noted that on free and personal plans, OpenAI, Anthropic, and Perplexity also offer the ability to opt out of sharing your data with the company (i.e., opt out of training on your data). Google and Microsoft respect the privacy of your existing account with the provider.

(4) API usage

OpenAI, Anthropic, Google, Microsoft, and Perplexity each specifically state they do not train on data sent through the API. This is the case for nearly every single AI provider.

I get the hesitation and paranoia accountants and firms have about sharing sensitive data. Hopefully the above information convinces you that the infrastructure of the AI providers is as safe as any other software in your tech stack. There are other risks, many of which are more complex than a typical accounting software provider. It’s still up to the firm to evaluate those risks and the security measures taken by the AI providers and make an informed decision.

Firms that want to use AI need to be purchasing teams, business, or enterprise plans for their employees. It’s important for data security and confidentiality.