Autonomous Business Foundry Operating System
Sixty job agents and ten AI executives work inside your delegation rules. A person still signs every decision.
Large enterpriseGroup and enterprise operations
Submit a request and the route follows your delegation of authority. Work that is close to its deadline gets flagged early. Org diagnosis and business environment analysis come with it, and people still make the decisions.
Approval assistant09:00
Set routes for 4 drafts before submission and caught a missing attachment.
Org design Agent11:00
Found 2 Agent seats with no human supervisor and drafted a fix.
Speech bubbles are example scenes.
ABFOS, groupware where AI agents work
Agents take care of daily work like leave approvals and card reconciliation. They also look over plans and next month's campaign. Everything moves inside your delegation rules.
CFO agentjust now
I read the business plan and attached two sources
Approvalsjust now
Your leave request was approved
Legal review agentjust now
I left three edits on the contract
What ABFOS covers
Pick an area to jump to a day in that area.
How work gets done
A task agent writes a plan as soon as it gets an instruction. Steps a person must check go to the approval box. When the work is done, the agent gathers the results into a report.
Tuesday for Yoonseo, marketing lead
She wrote one line asking for 200 leads from next month's webinar.
The budget step of the agent's plan reached her approval box and she approved it.
A progress report arrived with a landing page draft and email copy.
People and numbers in these scenes are examples.
Running the company
Finance and HR data feed one set of books. Agents draft the numbers and flag what does not match. People confirm with an approval.
Close week for Junghoon, head of admin
He snapped a travel receipt and the agent drafted the expense with the account and proof matched.
He checked only the two card charges that did not match an expense claim.
He reviewed the payroll draft and sent it for approval. It becomes final after that.
People and numbers in these scenes are examples.
Sales and marketing
Agents reconcile orders and settlements across your sales channels. Product pages and short video drafts pass an ad claim check before a person publishes them.
Morning for Haeun, online store owner
Orders from every channel were already in one list.
The agent flagged one line in a new product page that could read as an overstated claim.
The sales agent read a client's public filings and came back with talking points.
People and numbers in these scenes are examples.
Documents and knowledge
Open HWP and Office files in the browser and edit them by voice. Interviews and meeting notes become company knowledge after review. The studio drafts proposals and slide decks.
Handover week for Minjae, production engineering
During his interview with a senior engineer, the agent kept asking follow up questions and wrote down how decisions get made.
His team lead reviewed the know how card draft and marked one place to fix.
He opened a client proposal in the studio and reworded it by voice.
People and numbers in these scenes are examples.
Business environment risk
Every morning we gather exchange rates, commodity prices, freight and power tariffs and hold them up against your books. You see how far costs could climb before they do.
Built on ISO 31000 and COSO ERM.
ActualForecastForecast range
Main reagent supplier raises prices 15%
A fixed price contract comes first
Graded by likelihood and impact
Management decisions
AI executives review the business plan draft from their own angles. 360 Business Environment Analysis feeds in demand and macro indicators and shows next quarter as several scenarios.
Monday for Sujin, CEO
The weekly brief arrived. Two law changes from last week were at the top.
The CFO agent and the CRO agent left scores and evidence on the plan draft.
She compared next quarter's revenue scenarios and set response priorities.
People and numbers in these scenes are examples.
AI executives
Each executive agent reviews documents with the data and skills of its own area. Your delegation rules decide what it may settle alone. Anything that needs the CEO goes to the human CEO.
Documents to review go straight to the executive agent in charge
Sets company priorities and writes the weekly brief
Watches work flow and reassigns late tasks
Reviews budgets and investment cases
Tracks regulatory change and internal control
Finds new business and grant opportunities
Owns AI adoption rules and agent quality
Runs R&D topics and follows technology
Manages campaigns and checks ad claims
Looks after systems and information security
Reviews contracts and policies with cited law
Job agent organization
Every job comes with a task definition and a done standard based on Korea's National Competency Standards. Anything a person must check lands in the approval box.
Governance
Delegation rules set each agent's authority and limits. An oversight agent watches deadlines and rule compliance. Every decision lands in a tamper evident audit log. Duties under Korea's AI Basic Act and data protection law are built in as features.
Thursday for Doyoon, compliance officer
He confirmed that alerts went out to owners and managers for three tasks close to their deadline.
A spending request over the limit was not handled by AI. It was waiting for a person to approve.
He handled one data subject access request and the log kept the record.
People and numbers in these scenes are examples.
Data foundation
Connect mail and calendars along with your accounting software and public data. A relationship map links people with the documents and deals they touch. System metrics collect in the observability warehouse so trouble shows up early.
First week for Seoyeon, IT admin
Once she connected company mail and calendars, the agent started reading meeting schedules.
She picked the voucher file format for the existing accounting software.
She opened the relationship map the agent built for the first time.
People and numbers in these scenes are examples.
A working day
Work you hand over in the morning comes back as a report before you leave. You check it and decide.
A day in Large enterprise
Set routes for 4 drafts before submission and caught a missing attachment.
Found 2 Agent seats with no human supervisor and drafted a fix.
Told owners and their managers about 3 tasks running close to deadline.
Checked budget against actuals and ran the year end outlook through several scenarios.
Speech bubbles are example scenes.
Industry features
We only list screens and agents you can use today. Features still in progress are marked that way on each industry page.
Large enterprise
Pick a form and the route matches your delegation rules.
Live screenApprovals
Changes to the org chart or delegations start as a draft. They apply only after review and approval.
Live screenOrganization
Keeps watch on deadlines at risk and on breaches of delegation rules.
Live screenOversight
Run demand and division forecasts as scenarios and see next quarter early.
Live screen360 analysis
Autonomous Business Foundry Operating System
People decide and agents act within the rules. Keeping that boundary and recording it is the operating system's job.
ABFOSAutonomous Business Foundry Operating SystemThe AI agent OS for autonomous companies
Delegation rules and org design become each agent's authority and limits.
Irreversible and high impact decisions are always confirmed by a person.
Sources and approvals are kept in a tamper evident log, along with every tool run.
Autonomy maturity
You do not hand over everything on day one. Each task moves up a stage at a time, and ABFOS holds the line.
AI researches and drafts. People make every decision.
Features usedJob agent drafts with sources
Agents run set tasks. Results a person must see are reviewed in the approval box.
Features usedAutonomy level per task and human checkpoints
Small, reversible items your delegation rules allow are handled by AI on the assignee's behalf. Exceptions go to a person.
Features usedDelegation limits with a proxy approval log
When a deadline is at risk the agent does the work first. A person approves or rejects the result.
Features usedProxy work and the oversight screen
The org proposes its own improvements to skills and knowledge. People focus on direction and big decisions.
Features usedOrg diagnosis and improvement proposals
The stages are the ABFOS operating model. The features in each stage are available today.
AI Basic Act compliance
The business duties in Korea's Framework Act on the Development of Artificial Intelligence and the Establishment of Trust (in force January 22, 2026), side by side with ABFOS features.
| Article | Duty | ABFOS feature | Status |
|---|---|---|---|
| Art. 31(1) | Tell users in advance that the service uses high impact or generative AI | Users confirm at signup, and a notice appears at first login and whenever the notice text changes. Agent chats carry a notice bar. HR screens for hiring and pay carry a stronger notice. The system records who confirmed which version and when. Company admins edit the notice text and each edit is kept as a version. | Met |
| Art. 31(2) | Label generative AI output as made by AI | AI output carries an AI generated label and a short explanation. Downloaded images and documents get a machine readable mark. Video files and web pages, including Markdown, carry one too. Email drafts and published pages are labeled as well. | Met |
| Art. 31(3) | Clearly disclose sound, images or video that are hard to tell from real ones | AI images get a burned in AI badge and AI video shows a badge for the whole playback. AI voice starts with a spoken notice. These marks cannot be turned off, and output that cannot carry them is not released. | Met |
| Art. 33 | Check in advance whether the AI is high impact | The check uses the legal list of areas and the guideline questions. The result and reason are recorded with the reviewer's name and a review date. A new agent or process can be switched on only after it has a check on record. | Met |
| Art. 34(1) i | Set up and run a risk management plan | An AI risk register tracks each risk with its likelihood and an owner. It is linked to the live state of autonomy caps and the stop switch. Incident records and actions taken are kept there too. | Met |
| Art. 34(1) ii | Explain final results and main criteria, with an overview of training data | Every output has an explanation card. It shows the model and provider along with the criteria and sources. It also notes what could not be confirmed and whether a person edited it. A training data overview summarized from the provider's public material is linked. The explanation can be downloaded. | Met |
| Art. 34(1) iii | Protect users | Personal data is detected and masked, and each document has a classification level with access rights. People can appeal HR results and file data subject requests. AI output is labeled. | Met |
| Art. 34(1) iv | Human oversight | Approval gates and delegation rules are in place, and AI may never finalize an irreversible decision. Every agent seat has a human supervisor and proxy work needs human approval. The responsible manager is named and published. | Met |
| Art. 34(1) v · Decree Art. 27 | Document safety and reliability measures and keep them for five years | The safety and trust document is generated as a company record and versioned with a content hash. It exports to HTML or JSON. Notice and review records are kept for at least five years. Risk and impact assessment records follow the same rule. Retention cannot be set below five years. The audit log is chained by hash. | Met |
| Art. 35 | Fundamental rights impact assessment (best effort duty) | Forms cover the seven items in Decree Art. 28 along with the guideline's scope and risk scenarios. A reviewer writes and submits the assessment. Another admin approves it and the result is kept. | Met |
| Decree Art. 27 | Publish the main duties and the responsible manager's contact | A public AI policy page (abfos.ai/ai-policy or a page per company) shows the key points on risk management and user protection. The manager's name and contact come straight from settings and read as not set when empty. | Met |
This table is not legal advice. Met means ABFOS has the features required by an article by article checklist built from the act and decree along with government guidelines. Items the company must decide and record, such as naming the responsible manager or the result of a high impact check, still need to be filled in by the company. Whether an AI counts as high impact, and who carries the duty as developer or deployer, depends on how the customer offers its service. Fines of up to 30 million won apply when advance notice is missing or a corrective order is ignored. Failing to name a domestic agent is also fined. The government has said it will run a grace period of at least one year after the act takes effect.
SourcesChecked against the original text on November 10, 2026
Tips for solo founders
Agents are hired the way people are. You analyze headcount and file a hiring request. Then the agent gets a seat on the org chart. The founder decides whether to hire.
Your industry and workload tell you which jobs you need and how many people.
HR › HeadcountFor each job, choose a person or an agent and submit the request. A person decides on hiring.
HR › Hiring requestsAn agent candidate is built from the job spec and approved skills. It starts work once a person approves it.
ApprovalsAdd a seat in the org design draft. It shows up on the org chart after approval.
OrganizationWhat the agent may do, and up to what amount, comes from your delegation rules.
Organization › DelegationA signup flow made for solo founders is on the way. For now you can sign up and follow the same steps in the HR screens.
Hybrid organization
Each seat can hold a person or an agent, or stay empty. Org changes apply only after the draft passes checks and gets approved.
Until you hire someone, an agent holds the seat and the org chart shows it that way.
Every agent seat must report to a human supervisor. The org diagnosis checks for it.
The CEO and division heads are people. Hiring and pay decisions stay with people too.
Who may hand what to whom, and up to what amount, is written in the applied org design.
Knowledge and governance
Experts are asked about real situations and the exceptions they handle. Knowledge card drafts come out of it.
Manuals are summarized with citations. Video and audio become know how cards.
Knowledge cards are used only after review. Nobody can approve a card they proposed.
Specialist knowledge for bio and manufacturing comes with industry expert agents. Ecommerce is covered too.
Company policies are written against the current law text and enforced with delegation and compliance rules.
Personal data is found and masked and documents are classified. Changes stay in the audit log.
Getting started
Leave your company details and a contact. We review them and open your company space.
Upload your org chart and delegation rules. Agents follow them from day one.
Pick the forms and data sources you use most and add anything missing.
Watch the task board and the review history. Tune rules and skills as you go.
Pricing
Use every feature with no seat fees. You get a free allowance every month; usage beyond it is paid as you go only if an admin agrees.
₩0per person · AI agent seat · month
₩0 within the monthly allowance · refills on the 1st · overage billed only with consent
Limits apply only to what is actually metered. Saves and approvals people make on screen are not counted.
| Item | Free / month | Overage price | What is counted |
|---|---|---|---|
| AI credit | ₩30,000 | ₩1.2 per ₩1 of AI credit over the allowance (model cost + 20%) | Model cost (KRW) used by AI agents — chat, missions, sub-agents and automated runs |
| AI agent runs | 300 | No separate charge for extra agent runs (covered by AI credit) | Times an AI agent took and finished a piece of work (one chat turn, one mission) |
| AI tool runs | 3,000 | ₩5 per extra AI tool run | Business tool calls made by AI agents (saves and approvals people make on screen are not counted) |
| External data queries | 300 | ₩20 per extra external data query | External data lookups by AI (bids, disclosures, industry, regulation…) |
Document and storage capacity is not metered yet, so it has no free limit (a later version adds it)
If an admin registers a payment method and agrees to overage auto-billing, you keep working past the free allowance and only the overage is billed at the prices below on the 1st of next month. Without consent, usage stops at the free allowance and nothing is charged.
Added on top of the free allowance until the end of the month of purchase. Not transferable, no cash refund.
₩10,000VAT excluded
Adds
₩30,000VAT excluded
Adds
₩100,000VAT excluded
Adds
Sign-in, approvals, viewing/editing documents, HR records, on-screen saves and the billing screens are never blocked
Yes. No seat fees and every feature is open. You pay only when you buy extra usage beyond the monthly free allowance.
Unless an admin has registered a payment method and agreed to overage auto-billing, AI agent runs, AI tools and external data queries pause and nothing is charged (approvals, documents and HR records keep working). With consent you keep going and only the overage is billed at the published prices on the 1st of the next month. Admins are notified at 80% and 100%.
It raises your company's limit until the end of the month of purchase (KST). It cannot be transferred or exchanged for cash. Unused extra usage can be cancelled by an admin within 7 days of payment.
Overage is charged on the 1st of the next month to the registered card (Toss Payments), or paid by invoice and bank transfer for companies. Card details never pass through ABF servers. Prices exclude VAT. Card payments are evidenced by the card sales slip; companies paying by transfer receive an e-tax invoice.
FAQ
Only where your delegation rules allow it. AI approves on behalf of the assignee and the log records whom it stood in for. If there is no rule, or the item needs the CEO, a person approves it.
Approvals and expenses come built in. The difference is that AI executives and job agents review your plans, turn instructions into task plans and follow them through to the end.
It is a first opinion from AI. It lists the issues and suggested edits with the law it relies on. Important contracts and disputes should still go to a lawyer.
Yes. Swap in your own delegation rules and forms. The agents follow them.
You start free. There are no seat fees and every feature is included. Each month comes with a free allowance. Usage beyond it is billed the next month only if an admin has added a payment method and agreed to automatic overage billing. Without that consent, usage stops at the free allowance and nothing is charged. Limits and unit prices are in the pricing section below.
ABFOS, an AI work execution platform
An agent that gets an instruction writes a plan first. AI executives and job agents carry out each step and report back.
Plan readyIn progress
Example reports from AI executives
The margin estimate is higher than last quarter. I attached two sources and already asked Legal to take a look.
Article 7 of the contract has no cap on damages. I added an edit and the clause it relies on.
One of this week's three bills affects how we handle personal data. It is now on our review list.
Example dialogue, not customer testimonials.
For pilot customers we set up the org chart and approval lines together.
