We get a firsthand look into organizations doing millions to tens of billions in revenue. And while AI usage is up, outcomes are sporadic. The ads you see aren’t telling the whole truth. Few teams we meet are scaling capacity with those agents, which is how we define “agentic”—not just bots taking action, but agents working so effectively, they unlock more time, money, and options.
By that measure, lots of agentic finance setups do not yet pass. Hence, we are on a mission to clarify what agentic finance really means at the technical edge with clear use cases that reduce the hype.
One of the best things we can offer is to help you study the tools. For example, should you be using Anthropic’s Claude Cowork or Lensing’s AiSpecify? Or both together? I’ll give you the quick answer, then we’ll unpack it in depth: Use both.
Claude Cowork is a generalist agent that can operate your desktop, sort of like when IT remote-controls into your laptop. It is useful. It also lacks guardrails and can wantonly delete critical data. Whereas AiSpecify is a bigger tool with a narrower use: building, modifying, and maintaining business applications like your ERP. If you need to create an app to run your close process, use AiSpecify. And if you need an agent to run that app, use Claude Cowork.
Let’s go into the details on how they differ and work together.
What is Claude Cowork?
Claude Cowork is an add-on product from Anthropic that gives the LLM the power to take a series of actions on its own, on your desktop. You specify its level of access. For example, just one single folder, and it can open files, transform CSV files, or review documents to generate a strategy.
Early reviews say it’s effective. Zarina Zayas-Ortiz at Terabase Energy used it to write a strategy for her chief operations officer to review, and they concluded its approach was sound. It was a big leap in improvement over Microsoft Copilot, which couldn’t work a CSV.
What is AiSpecify?
AiSpecify is an AI tool for editing your business software, especially accounting and finance applications like your ERP. What that means in practice is, a non-technical user can chat with AiSpecify to say, “We’re tired of running our deal desk out of Google Sheets, build me a module to accomplish this within our ERP,” and it will do it.
This is a remarkable innovation. And the software AiSpecify produces actually works and is maintainable. Unlike generalist AI models like Claude, ChatGPT, or Gemini, AiSpecify is a software for building business software. Whereas those other models might do things differently each time, AiSpecify has a very rigid process. This allows it to do something no AI tool has before, which is to understand the full context of your chart of accounts, procure-to-pay workflows, order-to-cash flow, reports, and the whole setup in your existing business systems. Like a responsible coworker, it knows your business.
There are three reasons AiSpecify is able to do this, where LLM tools can’t:
1. AiSpecify uses spec-driven development
When you ask most LLMs to generate software, or specifications for software, the process is ad hoc and often changes. They also want to move fast and try to satisfy your request as quickly as possible. This means they’ll build the whole thing at once. This leads to an effect we call “80% thrill and 20% chill”—the first 80% of building that app or feature is thrillingly fast. It looks nearly complete. But getting it to function with your ERP or billing system, or move a button, or fix a left-aligned figure is excruciating—in the final 20%, a chill sets in where you realize it’s actually not possible.
AiSpecify works differently. It has a hardwired software development process that mirrors what developers and system architects do: Follow the software development lifecycle. It breaks that building process into discrete stages. At each stage, it invites your input, helps you research, and guides you to build a specification that outlines precisely what you want that app or feature to do. Then, it builds a little more and asks, “How is this?” Step by step, it will generate an increasingly realistic software interface you can test.
In this way, you iteratively build toward the exact software interface and function that you need to actually use it reliably at work.
2. AiSpecify produces maintainable software
The reason Microsoft’s workflow vibe-coding tool Power Apps helped hundreds of thousands of people produce millions of applications and yet hasn’t led to a single application you have heard of is simple: Those apps didn’t work. They had no mechanism for understanding the full context of the enterprise systems they were meant to function within and amongst. Even the people vibe-coding them didn’t have enough context. With all the configurations and data flows, it’s just so massively complex. And a generalist LLM like Copilot, which is behind Power Apps, can only operate on guesswork—and leave security gaps.
AiSpecify’s inventors solved this by designing the system around a full-featured ERP platform. In fact, they built it to help them build that ERP application. Franz Faerber and Sandeep Chopra spent five full years in stealth with no pressure to commercialize, and crafted an AI tool to make that build go faster—they found that by using it to generate specifications iteratively, there were fewer issues. And it took weeks rather than months to build nearly functional prototypes. They then realized that if it was helpful to them, it’d be helpful to others.
This enterprise app origin means AiSpecify is an app, not just an agent—a full-fledged software that executes a workflow. And it’s built to understand the data models and architecture of financial systems. So when you ask it to build a deal desk app for you, it isn’t guessing. It’s drawing upon an actual repository of deal desk applications. In other words, it knows.
And further, because you build an app or feature iteratively with AiSpecify, and it does the coding, you can edit the specification to edit the code. When you change the requirements, it changes the app—and captures all those changes for a full audit trail. Which means non-technical users can edit their ERP and maintain those changes, no tech team required.
3. AiSpecify can replace your aging business systems
Related to the above point, you can also ask AiSpecify to reverse-engineer your legacy work applications, and rebuild them in newer languages for better performance. Here’s a real scenario: One team that was migrating from Oracle to SAP wanted to bring in an old, custom application into SAP, but nobody knew how it worked anymore. They’d delegated that task to an outsourced team in India which had spent one year and made little progress. When AiSpecify emerged, they had an internal team give it a try. The tool reviewed all 30 structured technical documents and helped them rewrite and migrate the application in a few weeks.
In this way, AiSpecify allows non-technical business users to modify their ERP and business systems in ways that previously weren’t feasible.
Claude Cowork pros and cons
The pros of Claude Cowork are many: It’s one of the more reliable generalist agentic LLM tools on the market, and it navigates your desktop and most applications, including a browser without getting stuck.
The downside is Claude Cowork sometimes vastly overcomplicates the work, and spawns sub-agents or creates ephemeral applications to perform processes that it could more easily have done on your own. It works by brute force at times. It is also not auditable. This is a dealbreaker for many larger organizations, because LLMs are known to complete tasks by simply deleting them. For instance, helping you reach inbox zero by the fastest possible method.
AiSpecify pros and cons
The upside to AiSpecify is that it is the only enterprise-ready AI tool that can build ERP apps and features. No other AI startup or ERP provider offers this. And AiSpecify’s history (five years of development upon a full-featured ERP platform) is the best assurance you have that the system will understand your business, act responsibly toward it, and build apps that are safe and maintainable.
AiSpecify is also under active development and is being run by companies ranging from small startups up to some of the biggest financial companies in the world, which means it’s being rigorously pressure-tested and approved. It’s fully auditable—its reasoning and state are captured in the specification—and it can be used collaboratively by a distributed team.
The downside to AiSpecify is the intentional narrowness of its use case: It’s to build applications. You can’t run it like a general purpose agent. But that’s what the generalist models are for.
When should I use Claude Cowork?
Use Claude Cowork for generalist tasks like merging CSV files or running variance analysis. The clearer the process, and the better you can define the steps and inspection process, the better. This means Cowork excels at rote, repetitive tasks of exactly the kind of fast-moving executives like Zarina Zayas-Ortiz intend to hand off.
When should I use AiSpecify?
Use AiSpecify for developing apps and features in your business systems like your ERP. Because it understands your operations, you can use it to build new apps, consolidate unnecessary old ones, or maintain old software with fewer technical resources. Companies are using it to build deal desk software, treasury management apps, creating a customer onboarding software, supply chain apps, procurement apps, and much more.
When it comes to enterprise resource planning, what should I keep in mind about AI agents and applications?
You should keep in mind that there’s really two parts to any modern finance stack: the deterministic core and the non-deterministic AI layer on top of it.
The deterministic core is the ERP foundation that defines your most critical processes like record-to-report, quote-to-cash, procure-to-pay, and so on. Those must run reliably and without error or innovation. You have to trust that it’s all been handled as you’ve laid out. This part of your stack does as it’s told.
The other part of your stack is bringing AI intelligence into the mix to resolve edge cases or launch agents that handle repetitive tasks. For example, using AI to conduct three-way PO matching or using an agent to trial-run your close and identify just the areas that require human intervention.
So, what’s the verdict?
Use Claude Cowork for repetitive, well-defined tasks with clear steps, and where Claude Cowork will have access to just the folders and applications you greenlight. Use AiSpecify for editing, improving, or creating business applications. It’s how your small, non-technical team can act like you have a team of ERP architects on staff, and create systems that agents can run for you.
When you combine an agentic tool like Cowork and an enterprise app-building software like AiSpecify, you become one of the few finance teams out there completing real work. Which is to say, saving your team time, not just consuming it.
