Peter Zhu is Co-Founder of Minervaan AI accounting firm backed by Y Combinator. Peter has been building AI solutions since the age of 15.
For the last two decades, SaaS has dominated the technology landscape. Founders crowned the insights that could lead to feature-driven products and raced to sell subscriptions as software ate the world. But today, AI is rewriting that playbook. Instead of delivering features, AI delivers service, capable of executing judgment-driven, high-variance tasks once reserved for humans.
I’m the founder of Minerva, an AI accounting startup. When telling people we’re funded by Y Combinator and other Silicon Valley investors, many anticipate that we’re building AI software and distributing it to businesses or accounting firms. The truth is, we’re nothing like that.
We’re an accounting firm. You read that correctly—we’re a services company, and our accountants close the books and file taxes for our customers the same way any other accounting firm would. The difference is that our AI agents automate the repetitive, menial work for our accountants so they can focus their energy on resolving complex issues and engaging with our customers.
In this article, I explain how to effectively combine human service with AI to achieve higher efficacy and outline some ideas on how business owners can capture value with this new paradigm shift.
Where AI Meets Service: The Opportunity And The Risk
The sweet spot for AI today is in tasks still performed manually because they’re too complex, too specific or too edge-case heavy to be automated through traditional code. Think accounting reconciliations, contract administration or compliance workflows—services typically performed by professionals.
Yet, challenges remain. AI makes mistakes. Hallucinations in large language models are well-documented. In mission-critical areas where accuracy outweighs cost or time savings, blind automation introduces risk.
The answer isn’t a “human-in-the-loop” approach in the old sense of manual checkpoints. Instead, the opportunity is AI-empowered humans. Just as students use AI to accelerate exam prep and developers use AI to scaffold codebases, professionals across industries can use AI to amplify their output.
In accounting, for instance, staff trained on a custom AI platform could cut reconciliation time from hours to minutes. In law, paralegals could draft contracts with AI assistance, leaving attorneys to refine high-value details. Rather than replacement, this is augmentation—a hybrid model where AI handles scale and humans handle sensitivity.
How Leaders Can Capture Value Now
Organizations looking to adopt AI as a service should focus on three priorities:
1. Identify rote workflows.
Look for high-friction, context-specific tasks where AI can reduce cycle time. For example, does your customer support team get blocked with simple questions that your QA software can’t currently automate because it requires searching up a specific detail in that customer’s database? This is a scenario where you can build an agent to respond to the request using the context-specific information from your customers.
We experienced this with one of our customers whose bank statements bundled up all of their ACH transfers. Tens of ACH transactions would be bundled up into one line and aggregated, making it difficult to process AP and perform reconciliation. Historically, an accountant had to find the file containing all ACH transfers and manually add numbers by trial and error to find which transactions correspond to which bundle. With AI, we built out a workflow to do this automatically, converting about 10 hours of manual labor per month into 30 seconds of compute time.
2. Build domain-specific AI platforms.
General models are powerful, but tuning for your industry improves reliability and reduces risk. Baseline AI models are great as generic assistants, but there’s a lot of room for improvement by prompt engineering the models with the right instructions, or even taking it one step further and post-training an open-source model with thousands of samples of domain-specific data.
Because we deal with a lot of numbers, we want to ensure our AI systems don’t rely on probabilistically generating sums and instead write code to execute precise mathematical computations. We do this by building out a separate “compute agent” module that we then prompt our agent orchestrator to call when it needs to add together a string of numbers.
3. Train staff as AI operators.
One of the most overlooked areas of value is having AI-native staff. With existing publicly available AI tools like Claude and Lindy, your team can get a ton of improvement just from incorporating these tools in their workflow.
This is why, at Minerva, we focus on training our staff as AI operators as much as we spend time developing AI tools to fill the gaps. The future is still very much human—but it lies in the hands of those who can leverage AI.
The Future Of AI As A Service
The SaaS era was about building and selling features. The AI era is about delivering service at scale. Companies that embrace this shift—by pairing AI with empowered human teams—can unlock new productivity curves without sacrificing accountability.
We aren’t in a world where machines can replace human judgment wholesale. But we are in a world where AI can amplify human service delivery in ways software never could. The winners will be those who stop thinking of AI as just another feature and start building businesses where service is the product and AI is the engine.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?
