Automating accounting close, bookkeeping workflows, and reconciliations

Excited to show every accountant how Puzzle AI Close works. This is the first agent builder built into the general ledger for accounting firms. Accountants are loving it.

Accounting software made reporting better.

But it never fixed the work.

The month-end close is still the same grind it was 20 years ago. Same manual steps, checklist, and late nights reconciling the same accounts every single month.

Here's how AI Close works.

Describe a task the way you'd hand it to a teammate.

"Reconcile this bank account."
"Categorize these transactions."
"Draft the accruals for this client."

Puzzle turns that into an AI agent that drafts the work for your review. Not AI that gives you insights. Not a chatbot that answers questions. An agent that does the actual work and puts it in front of you to approve.

AI does the work. The accountant makes the call.

That's why we built this into the ledger instead of bolting it on top of someone else's. Agents need to live where the data lives. Inside the general ledger. Not outside it, looking in through a one-way API.

AI Close is now available to every firm.

I recorded a quick walkthrough on why we built it this way and how it actually works for your firm.

AI can never replace accountants. This is why we built AI Close for Accountants and Firms as part of the Puzzle AI Suite. Every AI accounting tool takes the accountant out of the loop. But we put them back in.

AI Close is built for the month-end close, with the accountant in full control every step of the way.

Agents draft the work.

You approve every move.

Nothing touches your books until you say so.

Your Judgment. At AI Speed.

BlackLine announced on June 25 the expansion of the Agentic Financial Operations Platform with new governance and observability capabilities designed to help finance teams manage AI agents without losing control of the financial record. It includes a preview program for Finance Control Console, a centralized command center for AI-powered financial operations. The console is designed to give finance leaders visibility, policy enforcement, risk monitoring, and audit-ready records across BlackLine-native, partner, customer-developed, and third-party AI agents.

Trintech Flux agent automates account fluctuation analysis during the close; Trintech Variance Analysis agent accelerates budget-to-actual performance review after the close — both operating within governed financial workflows

Pleo has launched a suite of AI agents to automate spend management, AP, and accounting workflows, allowing finance teams to focus on strategic tasks. Beta testing begins in July 2026.

Read: https://t.co/TOzMEd9zZZ

#Pleo #Fintech #AI #AgenticAI #SpendManagement

Using AI in accounting feels like working with Einstein — except you have to constantly teach Einstein what accounting is.

A little over 2 months into an AI startup as a Controller, I am having more fun than ever automating accounting workflows so I don’t have to do the work myself. I blew through my session limits 3 days in a row this week (consecutively mind you - 11am to 6pm, 6pm to 11am, rinse and repeat) - but I created financial reporting packs in days that would have taken an entire team weeks to prepare and review.

I no longer want to prepare any workpapers in Excel - that’s my agent’s job now. I am only going to review my agent’s work. Anything in Excel can be automated - no exceptions. In fact, this probably applies to the entire Office suite.

With that said, I am learning that the bottleneck is becoming… me. We have gone full circle - I remember back in my audit days waiting for CaseWear IDEA to finish selecting samples so I can begin testing, or waiting for Blackline to batch process reconciliations so I can start preparing reports. But now, I find myself struggling to keep up with all the different sessions and agents that are running in my Claude. Jumping between sessions to review AI output is harder than expected because of constant iterations. You add jumping in and out of meetings on top and it very quickly becomes absolute ADHD level chaos.

Because the AI works SO FAST, you need to speed up your review frequency as well, and constantly iterate to achieve the desired output. Gone are the leave-a-comment-and-wait-for-your-team review cycles - AI does 10 cycles in 1 minute.

Am I the only one hitting this attention span wall?

First, Anthropic didn’t ship another chatbot for bankers.

→ It shipped a full AI agent platform that sits above every data provider, every Microsoft Office app, and every internal system a financial institution runs, orchestrating end-to-end workflows that used to require teams of junior analysts and weeks of calendar time.

→ 10 production-grade AI agent templates covering everything from pitchbook assembly to month-end close to KYC screening.

→ Native Excel, PowerPoint, and Word integration.

→ Apache 2.0 on GitHub.

→ Citadel, Goldman, BNY, Carlyle, and AIG are already deployed.

And the real distribution play - an infrastructure partnership with FIS $FIS, which processes core banking for institutions handling roughly 12% of the global economy - means thousands of banks will get Claude’s agents without ever signing an Anthropic contract.

Combined with a $1.5B joint venture announced the day before with Blackstone, Goldman, and Hellman & Friedman, Anthropic now has three simultaneous paths into finance:

(1) direct enterprise, (2) infrastructure embedding, and (3) a PE-backed services arm reaching hundreds of mid-market firms.

The architecture, the partnerships, and the market reaction all point to the same conclusion: the value in financial services is migrating from the data layer and the workflow tool layer to the intelligence layer above both.

Anthropic intends to own that layer.

Deeper dive into what it means for data incumbents, fintech startups building in the space, and the firms racing to adopt, or trying to figure out whether they’re already too late is here:

Anthropic shipped ten ready-to-run agent templates for financial services. Not a preview- these are reference architectures you can install today. (Catching up on this one from last week, worth a second look)

The Ten: Research & coverage- Pitch builder. Meeting preparer. Earnings reviewer. Model builder. Market researcher. Finance & ops- Valuation reviewer. GL reconciler. Month-end closer. Statement auditor. KYC screener.

Each runs as a plugin alongside the analyst, or as a Managed Agent running autonomously with full audit logs.

Three things stood out:
1. Packaging is the product. Skills + connectors + subagents, pre-stitched. Days, not months.
2. Context travels across M365. Excel → PowerPoint → Word → Outlook, no re-explaining.
3. Data layer is the moat. New connectors- Dun & Bradstreet, Guidepoint, Third Bridge, Verisk, IBISWorld, plus a Moody's MCP app on 600M+ companies - on top of FactSet, S&P Capital IQ, PitchBook, Morningstar.

Opus 4.7 leads Vals AI's Finance Agent benchmark at 64.37%. Walleye Capital reports 100% Claude Code adoption across 400 employees.

If you run a finance or coverage function- pick one agent (Pitch builder or Month-end closer), point it at one real workflow this week.

Announcement: https://lnkd.in/gAtRiTRE

We're releasing ten new Cowork and Claude Code plugins, integrations with the Microsoft 365 suite, new connectors, and an MCP app for financial services and insurance organizations.

80% of finance teams blame their slow close on waiting for data — even after investing in automation.
(Source: CFO Dive, May 2026)

Meanwhile, teams using AI in their close are cutting cycle time by 40-50% and reconciliation errors by 90%.
I used to be in the 80%. Two things got me out: process redesign + AI in the right places 👇
⸻
🎯 First: Fix WHEN you work, not HOW FAST
Stop treating month-end as a 10-day sprint. Spread reconciliation across the month, bank, AP, AR, intercompany — weekly. By Day 1, you're confirming, not discovering.
The industry calls this Continuous Close. It's the biggest unlock for lean teams.
⸻
🤖 Then: Layer AI where it actually saves hours
Here are 3 things I now let AI handle — with exact examples:
Variance Analysis (2 hrs → 15 min)

I export the P&L to CSV and paste into Claude with this prompt:
"Here's actuals vs budget. Flag every line item with >±10% variance AND >$50K impact. For each, tell me: is this seasonal, one-time, or a trend shift? Draft a 2-sentence explanation."

AI returns a flagged, prioritized list with draft commentary in 30 seconds. I spend 15 minutes reviewing and editing — not 2 hours building it from scratch.
Board Commentary (2 hrs → 20 min)

After variance review, I prompt:
"Based on these 5 flagged variances, draft a 150-word management commentary. Tone: factual, forward-looking. Highlight the 3 biggest movements and their business drivers."

AI writes the first draft. I add the context only I know — market shifts, pipeline changes, operational decisions. 20 minutes total.
Anomaly Detection (manual review → exception-only)

Instead of reviewing every journal entry, I prompt:
"Here are this month's journal entries. Flag anything that looks unusual — duplicate amounts, round-number entries over $10K, entries posted outside business hours, or accounts that don't normally have activity."

AI catches what human eyes miss at 2am on Day 3. I review exceptions only.
⸻
⚡ Two process hacks that cost nothing:
🚫 Hard cut-off, Day -2: Departments submit accruals 2 days before month-end. Miss it? Next month. I put a "Late Submissions" line in the CEO report. Nobody appears twice.
📝 Same-hour commentary: Write variance notes the MOMENT P&L closes — not 48 hours later. Delayed commentary = fiction. Same-hour = insight.
⸻
Results:
📉 Close: 10 days → 4
🤖 Variance work: 2 hrs → 15 min
📝 Board commentary: 2 hrs → 20 min
🎯 Board pack: Day 4 instead of Day 12
⸻
🛠️ Tools worth exploring: → FloQast / Numeric — close task management → HighRadius — AI anomaly detection at scale → ChatGPT / Claude — variance, commentary, anomaly flagging (free)

The secret? AI doesn't fix a broken process. Fix the WHEN first. Then let AI handle the grunt work.
Which part of your close eats the most time? Drop it below 👇

#CFO #MonthEndClose #AIinFinance #FinanceLeadership #StartupCFO #ContinuousClose #FinanceAutomation #CFOInsights