#llm
Llm
Large language models for accounting automation and insights
Bitwave Just Open-Sourced 'Agentic Finance': What Happens When AI Agents Stop Using Your Accounting Software Like a Human Would
Bitwave open-sourced an agent-native accounting CLI and MCP server on July 23, 2026, arguing AI agents need command-line, diffable interfaces rather than GUI dashboards — the same properties git-tracked plain-text accounting has offered since before "agentic finance" was a marketing term.
Xero's Claude Integration: What Small Business Owners Should Know Before Connecting Their Books
On May 12, 2026, Xero went live with an Anthropic Claude integration that lets 4.5 million subscribers query live invoices, bank transactions, and reports conversationally. Here is how the bidirectional connection works, what the JAX Assure guardrails and session-only data policy actually promise, a due-diligence checklist before granting OAuth access, and why a plain-text Beancount ledger gives any AI the same access with no integration at all.
Connecticut's CTDPA Now Covers Small Businesses: Neural Data, LLM Training Disclosures, and the July 2026 Rules
Connecticut's amended CTDPA took effect July 1, 2026, lowering the coverage threshold to 35,000 consumers, classifying neural data as sensitive, and requiring conspicuous disclosure of AI and LLM training on personal data. Processing any sensitive data — even one record — now triggers coverage, the 60-day cure period is gone, and penalties reach $5,000 per willful violation.
Agentic Commerce for Small Retailers: How to Get Your Store Recommended by AI Shopping Agents
AI-driven traffic to US retail sites rose 393% year over year in Q1 2026, and AI-referred shoppers now convert 42% better than other visitors. A practical guide for small retailers on structured product data, the ACP and UCP protocols, llms.txt, detecting AI agent traffic in server logs, and tracking agent-referred revenue by channel.
Generative Engine Optimization (GEO): How Small Businesses Get Cited by ChatGPT, Perplexity, and Google AI Overviews
Generative Engine Optimization (GEO) is how small businesses get named in AI answers from ChatGPT, Perplexity, and Google AI Overviews — a channel where over half of citations come from pages outside Google's top 10. This guide covers the seven retrieval signals AI engines use, a five-step checklist (Google Business Profile, schema markup, direct-answer content, third-party validation, NAP consistency), and the mistakes that quietly suppress citations.
LLM API Costs Are COGS, Not Overhead: A Gross Margin Guide for AI Wrappers
LLM inference is a variable cost of delivering your product, not software overhead — ICONIQ's 2026 data puts AI product gross margins at a 52% industry average, well below the 70-80% typical of traditional SaaS.
MiniMax FY2025 Earnings: $79M in Revenue, and a $1.9 Billion Loss That Isn't What It Looks Like
MiniMax's first annual report as a listed company shows FY2025 revenue of $79.0M (+158.9% YoY) and gross margin up to 25.4%, alongside an IFRS net loss of $1.87B — but $1.59B of that loss is a non-cash fair-value remeasurement of convertible redeemable preferred shares that converted to equity at the January 9, 2026 Hong Kong listing. Adjusted net loss grew only 2.7%, from $244.2M to $250.9M. A public Beancount ledger of FY2022–FY2025 shows exactly where the paper loss sits and why it disappears in FY2026.
Beancount.io's Ask AI Can Now Write Entries and Read Receipts
Beancount.io expanded its Ask AI assistant from a read-only Q&A tool into an action-capable agent that drafts and writes ledger entries, reads uploaded receipts and bank-statement screenshots, and commits nothing without explicit user approval — live at /agent.
Beancount MCP: Connect Your Ledger to Claude, Cursor, and Any AI Assistant
The Beancount MCP server connects your plain-text ledger to Claude, Cursor, Windsurf, and any MCP-compatible AI client over OAuth 2.1 — ask questions, run BQL queries, and commit ledger edits without leaving your AI tool.
California SB 53 Compliance: A Practical Guide to the Transparency in Frontier AI Act
California's SB 53 (Transparency in Frontier AI Act) took operative effect on January 1, 2026, requiring foundation model developers training above 10^26 FLOPs to publish safety frameworks, report critical incidents to Cal OES within 15 days (24 hours for imminent threats), maintain anonymous whistleblower channels, and face civil penalties up to $1 million per violation enforced by the California Attorney General.
The EU AI Act Lands on U.S. SaaS Companies This August: A Practical Compliance Guide
A practical guide for U.S. SaaS founders, foundation model providers, and AI agent builders navigating the EU AI Act before the August 2, 2026 enforcement deadline — covering Article 22 representatives, Article 50 transparency disclosures, GPAI Code of Practice obligations, fine tiers up to 7% of global turnover, and procurement-questionnaire readiness.
User Experience and Feedback on LLM-Assisted Plain Text Accounting
This article delves into user experiences with LLM-assisted plain text accounting, highlighting usability challenges, workflow bottlenecks, and the features users desire most in financial tools.