Amazon Web Services released two new tools this month because the company finally said out loud what accounting firm partners already know: AI agents don’t understand your business. They can write code, summarize emails, and draft content, but they can’t tell the difference between a client’s deferred revenue account and a prepaid expense. They don’t know which clients are on cash basis and which are accrual. They can’t see that one manufacturer depreciates equipment over seven years while another uses five.

AWS Context is the company’s answer. It’s a service that builds knowledge graphs from your corporate data so AI agents can see relationships, hierarchies, and business rules before they act. The second tool, AWS Agents Security, locks down what agents can touch. Both services acknowledge a gap that matters for accounting and bookkeeping firms: if an AI agent can’t read your client’s chart of accounts, understand their industry, or respect their compliance calendar, it’s just expensive autocomplete.

For firms running on tight margins during month-end close, onboarding three new clients at once, or trying to carve out time for advisory work, this gap isn’t academic. It’s the difference between an agent that saves 40 hours a month and one that creates cleanup work. Let’s walk through what business context means in practice, why it matters for the agents accounting firms actually need, and how to think about building it into your operations before you spend another dollar on AI subscriptions.

What business context means when you’re closing 30 sets of books

An AI agent without context can pull a trial balance. It can’t tell you why the balance is wrong. It sees numbers in columns. It doesn’t see that your manufacturing client always has a timing difference in raw materials inventory at month-end because their supplier invoices lag by five days. It doesn’t know that the law firm client books retainer revenue monthly but the contract says quarterly. It can’t flag that the construction client’s job costing looks off because two change orders from last month haven’t hit the system yet.

Business context is the layer of knowledge that turns data into decisions. For accounting firms, it includes client-specific chart structures, industry norms, recurring journal entries, compliance deadlines, and the judgment calls your senior staff make without thinking. When AWS talks about knowledge graphs, they mean a map of how these pieces connect. This account rolls into that category. This client follows this revenue recognition rule. This variance always happens in Q3 and here’s why.

Most firms store this context in three places: the partner’s head, a messy procedures manual that nobody updates, and the notes field in the practice management system. An AI agent can’t read any of those. So it reconciles accounts in isolation, misses patterns, and generates work that a junior accountant has to redo. The Omni Audit for accounting and bookkeeping starts by mapping where your firm’s context lives today and what an agent would need to see to do the work right the first time.

The agents accounting firms need and the context they’re missing

We build three categories of AI agents for accounting firms. Each one needs a different slice of business context to work. Let’s look at what they do and what breaks when the context isn’t there.

Month-End Close Agent

This agent pulls bank feeds, accounts payable, accounts receivable, and payroll data. It reconciles balances, flags variances above a threshold you set, drafts journal entries, and prepares a close pack for partner review. In a firm with 30 monthly clients, it should cut close time from 50 hours to 15.

Here’s what it needs to know: which accounts reconcile to external sources and which ones don’t, what variance threshold matters for each client (a $500 swing is noise for a $10M manufacturer but a red flag for a $400K retailer), recurring entries that post every month, and the order in which accounts close. Without that context, the agent reconciles everything, flags 200 variances, misses the three that matter, and forgets to accrue payroll taxes because nobody told it that step happens before the close pack goes out.

One firm we worked with tried a general-purpose AI tool for close. It reconciled bank accounts accurately but couldn’t handle intercompany eliminations because it didn’t know the client had three entities. The senior accountant spent four hours fixing entries the agent drafted. That’s not a time saver, it’s a time shift. The agent worked hard and delivered garbage.

Client Onboarding Agent

This agent collects documents from new clients through a guided workflow, sets up the chart of accounts based on industry and entity type, maps the client’s existing categories to your firm’s standard structure, and produces a clean opening trial balance. For firms that onboard five to ten new clients a quarter, it should compress a six-week process into ten days.

The context it needs: your firm’s standard chart templates by industry, the questions that determine entity structure (LLC, S-corp, partnership), the documents required for different client types (a construction client needs certified payroll records, a SaaS client needs deferred revenue schedules), and the cleanup steps that always happen (reclassifying personal expenses, splitting commingled accounts, correcting prior-period errors). Without that, the agent collects documents in the wrong order, sets up a chart that doesn’t match your workflow, and hands you a trial balance that needs two days of reconciliation before you can bill the first month.

We’ve seen firms lose 20% of new clients during onboarding because the process drags and the client doesn’t see value yet. An agent that understands your onboarding playbook turns that around. But if the agent doesn’t know the playbook exists, it just automates confusion.

Advisory Insights Agent

This agent reads each client’s monthly financials, compares performance to prior periods and industry benchmarks, surfaces three insights worth discussing, and drafts talking points for the partner’s advisory call. It’s designed to make advisory work scalable so the high-margin conversations don’t get crowded out by compliance.

The context it needs is the richest: what the client cares about (cash flow, margin, growth), what’s normal for their industry (a restaurant’s food cost should run 28-35%, a law firm’s realization rate should stay above 85%), the client’s goals (they want to sell in three years, they’re hiring two people next quarter, they’re refinancing debt), and the partner’s communication style (some clients want three bullet points, others want a full narrative). Without that context, the agent generates generic insights. Revenue is up 12%. Expenses are flat. Great, but so what?

The advisory agent is where business context makes or breaks ROI. A partner who gets useful talking points before every client call can handle 50% more advisory relationships. A partner who gets generic summaries ignores the agent and writes the talking points manually. You paid for the agent, you’re still doing the work.

Why AWS Context matters and what it doesn’t solve

AWS Context tries to solve the knowledge graph problem at scale. It connects to your data sources, maps relationships, and makes that map available to any AI agent you run. For accounting firms, that means you could theoretically feed it your practice management system, your chart templates, your client data, and your procedures, and the agent would see how it all connects.

Two things make this interesting. First, it’s infrastructure. You’re not buying another point solution that works one way for one task. You’re building a layer that makes all your agents smarter. Second, AWS is betting that business context is the next bottleneck. They’re right. Every firm we work with has agents that could save 30 hours a week if the agents just knew more about how the firm operates.

But AWS Context doesn’t build the knowledge graph for you. You still have to define what the agent needs to know, structure that knowledge so a machine can read it, and keep it updated when your processes change. For a firm that’s still storing procedures in a Word doc from 2019, that’s a bigger lift than buying a subscription. This is where the AI audit for accounting and bookkeeping becomes the forcing function. You can’t build context for an agent until you know what work the agent will do, what decisions it will make, and what knowledge it needs to make those decisions correctly.

What building context looks like in practice

Let’s take the Month-End Close Agent and walk through the context it needs for one client. Say you’ve got a $5M manufacturing client, accrual basis, monthly close, 40 accounts in the trial balance. Here’s what the agent has to know before it can close the books without supervision.

It needs the reconciliation map: which accounts tie to bank statements, which tie to sub-ledgers, which tie to nothing and just carry a balance forward. It needs variance thresholds: flag anything over $2,000 in cash, over $5,000 in AR, over $1,000 in inventory. It needs the recurring entry list: accrue $3,200 in payroll taxes on the last day of the month, book $1,800 in depreciation, defer $4,500 in prepaid insurance. It needs the close sequence: reconcile cash first, then AR, then inventory, then post recurring entries, then run the close pack.

That’s 12 pieces of structured knowledge for one client. Multiply by 30 clients and you’re documenting 360 pieces of context. If that sounds like a lot of work, it is. But here’s the thing: your senior accountants already know all of it. They just know it as muscle memory, not as documented process. Building context is really just writing down what your team already does so an agent can do it the same way.

The firms that get value from AI agents fast are the ones that pick one process, document the context for five clients, and deploy the agent on those five first. They learn what the agent misses, they tighten the context, and then they scale to the next 10 clients. The firms that fail are the ones that buy an agent, point it at all 50 clients, and expect it to figure out the context on its own. It won’t.

The security layer AWS didn’t have to announce

AWS Agents Security is the second tool Amazon released. It controls what data an agent can access, what actions it can take, and what approvals it needs before it does anything permanent. For accounting firms, this matters more than the context layer because one bad agent action can blow up a client relationship.

An agent that posts a journal entry without review and gets it wrong doesn’t just create cleanup work. It changes the client’s financials. If those financials went to a bank, a board, or the IRS, you’ve got a liability problem. If the agent pulled data from the wrong client file because it didn’t have access controls, you’ve got a confidentiality breach. If it overwrote a reconciliation that took your senior accountant three hours to build, you’ve got a morale problem.

Security for AI agents isn’t about hackers. It’s about guardrails. What can the agent read? What can it write? What requires a human to approve? For the Month-End Close Agent, a reasonable security model is: read all client data, write to a draft close pack, flag variances, draft journal entries, but don’t post anything until a partner reviews and approves. For the Client Onboarding Agent: read documents the client uploads, write to a staging chart of accounts, don’t touch the live client file until onboarding is marked complete.

Most firms don’t have a security model for AI agents because most firms don’t have AI agents yet. But if you’re planning to deploy agents that touch client data, the security conversation has to happen before the first agent goes live. This is one of the three outputs from an Omni Audit: a security and approval model that maps to your firm’s risk tolerance and client obligations.

The dollar math on getting this right

Let’s bring this back to the number that matters. A $3M accounting firm typically leaks $90K to $120K a year in operational waste. That’s time spent on work that doesn’t bill, doesn’t generate referrals, and doesn’t build the firm’s capabilities. Month-end close is the biggest single contributor. If your senior staff spend 50 hours a month closing books and 30 of those hours are reconciliation, data entry, and variance hunting, that’s 360 hours a year at a $150 loaded cost. That’s $54K in labor doing work an agent should handle.

Add client onboarding. If you onboard eight clients a year and each one takes 40 hours of setup, cleanup, and back-and-forth, that’s 320 hours. At the same loaded cost, that’s $48K. Add the advisory time you’re not billing because compliance crowds the calendar. If you could handle six more advisory clients a year at $1,500 a month, that’s $108K in revenue you’re leaving on the table.

The total opportunity is $210K for a firm at this scale. You won’t capture all of it with AI agents, but capturing half of it pays for the infrastructure, the audit, the agent build, and the first year of operation. The firms that get there are the ones that build context deliberately, deploy agents in controlled pilots, and measure the time savings in hours, not sentiment.

We’ve built a worksheet that maps the month-end close process for accounting firms, identifies where agents can compress time, and estimates the labor savings by client volume. You can download the Month-End AI Close Map for Accounting Firms and run the numbers for your firm. It takes 15 minutes to fill out and gives you a realistic range for what an agent could save.

What an Omni Audit finds and why it comes first

An Omni Audit is a 60-minute working session. You walk through one high-cost process in your firm with someone who’s built agents for accounting firms before. You map the process, identify the decision points, and document the context an agent would need to do the work. You walk out with three things: a process map, a context requirements document, and a security model.

The audit isn’t a sales pitch. It’s a forcing function. Most firms know they want AI agents but don’t know where to start. The audit picks the starting point, sizes the opportunity, and gives you a blueprint you can hand to a developer or use internally. For firms that decide to move forward with Omni, the audit becomes the design document for the agent build. For firms that want to build internally or use another vendor, the audit is still useful because the context work is the same no matter who writes the code.

Where this is heading

AWS Context is a signal, not a solution. The fact that Amazon is building infrastructure for business context means the market expects AI agents to move from generic tasks to specific workflows. For accounting firms, that’s good news. The agents you need aren’t the ones that summarize emails or draft blog posts. They’re the ones that close books, onboard clients, and surface advisory insights. Those agents require deep context, and the firms that build that context first will run leaner, bill higher, and grow faster than the ones waiting for a plug-and-play tool.

The gap between a useful agent and an expensive distraction is knowledge. Your firm already has the knowledge. The work is making it readable by a machine, structured enough to guide decisions, and secure enough to trust. That work isn’t optional anymore. The firms that do it now will have agents handling 40% of compliance work by the end of next year. The firms that wait will still be reconciling bank accounts manually while their competitors are billing advisory hours.

If you want to see what this looks like for your firm, start with the audit. If you want to understand the infrastructure, explore Omni Ops and see how we build agents that connect to your practice management system, read your client data, and respect your firm’s processes. If you want to learn how other firms are thinking about AI, the EDNA insights library has case studies, process breakdowns, and ROI models.

If you’re building with Claude or Codex right now, grab the free Working With Claude field guide. Thirty-two pages on the full ecosystem, Claude Code in depth, and how to roll agents out properly. Get the free guide.