Sep 1 – Oct 1, 2026 · 4 items
The AI week, for an accountant
A real edition, written for: Accountant
The one thing
Which of your recurring bookkeeping tasks could run unattended this quarter?
Dots launch: always-on assistants for ChatGPT
OpenAI launched Dots, always-on AI assistants that work without constant direction. They run on GPT-6 Astra with their own cloud computer. Dots connect to over 4,000 apps via plugins. They learn user preferences and work 24/7 on tasks. Users can inspect a dot’s work on its cloud computer. Dots can also use a user’s laptop with permission. They message progress or questions in ChatGPT, Slack, or Teams. Built-in safeguards include read-only tools for background work. Users control access, permissions, and approvals via Custom Rules. Specialist dots for organizations have their own identity and access. These are in focused enterprise pilots. Microsoft Agent 365 will integrate specialist dots. Dots roll out to Pro and Business Premium users in eligible markets.
Why it matters for you
Dots run 24/7 on tasks you assign. That matches your bookkeeping backlog.
- Pro or Business Premium plan required for access
- Connects to 4,000+ apps via plugins
- You control permissions and approvals
Try this
Set a Dot to reconcile a client’s monthly transactions
Paste this into your AI tool
Context: I am an accountant managing books for small business clients Task: reconcile this month’s transactions for [client name] against their bank statements Constraints: - use only the files I paste - flag any discrepancies I'll paste: [bank statement CSV] I'll paste: [transaction export CSV] Output: a reconciliation report with matched and unmatched items
Sources
Managed Deep Agents add user-level memory
LangChain released Managed Deep Agents 0.8. It adds user-owned credentials for per-user permissions. It adds user-level memory for caller-specific context. It adds HTTP channels for webhook-based agent access. It adds file transfer support in Slack. It includes a pre-built web search tool powered by Parallel. User memory is scoped to the authenticated caller. Agent memory remains shared across users. Default policies prevent user memory leakage in group channels. Connections support 23 services with managed OAuth. Slack workflows now accept files directly in conversations.
Why it matters for you
User memory keeps client-specific context separate. That stops cross-client mix-ups.
- User-owned credentials for per-user permissions
- User memory scoped to the authenticated caller
- Default policies prevent memory leakage
Try this
Ask your vendor whether user-level memory is on their roadmap
FrontierAgent open-sourced for local runs
ApodexAI open-sourced FrontierAgent, its agent framework. It offers a native command-line interface with a text-based UI. It supports ReAct and Agent Team workflow modes. It runs on macOS and Linux with one command. It requires no preinstallation and has no hard Docker dependency. The ReAct mode uses one stateful agent for research and file work. The Agent Team mode uses a coordinator to delegate tasks to parallel sub-agents. Both workflows share a sandboxed file system for inputs, workspace, and outputs. The framework, tools, workflows, and evaluation suite are modular and reusable independently.
Why it matters for you
FrontierAgent runs on your machine without Docker. That suits your small firm setup.
- Native command-line interface with text-based UI
- ReAct and Agent Team workflow modes
- Modular tools and evaluation suite
Try this
Test FrontierAgent on a sample client dataset
Paste this into your AI tool
Context: I am testing FrontierAgent for bookkeeping tasks Task: categorise these transactions for [client name] Constraints: - use only the categories in my chart of accounts - flag any items that need review I'll paste: [sample transactions CSV] I'll paste: [chart of accounts list] Output: categorised transactions with review flags
Sources
Hermes Agent refactored its own codebase
Nous Research used Hermes Agent to refactor its own codebase. It deployed 1,393 subagents over nineteen active hours. The refactor cut non-test Python by 34.4%. Model spend totaled roughly $19,300. A manual effort was estimated at $150k to $1.8M. The largest file shrank from 34,847 to 5,512 lines. Workers split giant files and consolidated duplicate helpers. Reviewers caught regressions missed by existing tests. Lookup costs for agents fell sharply after the refactor. The team updated Hermes skills with lessons from the run.
Why it matters for you
Hermes cut 34% of its code with agents. That shows what’s possible for your workflows.
- 1,393 subagents over 19 hours
- Manual effort estimated at $150k–$1.8M
- Largest file shrank from 34k to 5k lines
Try this
Review your most repetitive bookkeeping task for automation
Build this
Every week, one small thing to build with AI in something you actually care about. No work in it. Five minutes to set up, and worth keeping if it earns a second run.
A gift list for the people you already need to buy for, built from what you already know about them
Five minutes to set up
Context: I am making a gift list for people I already need to buy for Task: create a gift list using only the interests and preferences I already know Constraints: - no wishlists or hypotheticals - no purchases or new information I'll paste: [list of people and what I already know about each] Output: one gift idea per person, with a one-line reason tied to what I already know
- Have ready the list of people and what you already know about each
- Paste the prompt into ChatGPT
- Check that every gift ties to a fact you already have
- Save the list for when you next shop
Yours arrives Thursday.
This one was written for a accountant. Tell us what you do and the next one is written for you — same news, your job, once a week.