Product vision · October 2026
AgenticObjects’ thesis for agentic analytics in the new business world of the AI-native era.
The human on the other side has not changed since 1981.
Since personal computers began to spread in 1981, companies have produced more information every year, and agents now make producing more of it almost free. So the answer is not more output: it is records a person can check, trust and act on, with the evidence for every figure one click away.
Ali Yıldırım, Founder and CEO, D-CAT Group
A reader’s attention
Finding
Recommendation
Brief
Live today
We are not building a new data world.
Agents sit on the structure your company has invested in for years. They read the numbers it already agrees on, and they leave behind records that keep their evidence.
Operational systems
ERP, sales, finance: the transaction data the company runs on.
1Semantic model
Warehouse and tabular models: the numbers the company trusts and agrees on.
AgenticObjects agents
Read-only access. They never write to customer systems and take no action.
Agentic objects
Finding · Recommendation · Brief: persistent records bound to their evidence.
What makes a record defensible
Every figure has a structural address. Its origin can be traced to a value in a sealed query result. The anatomy of a record
The model never computes a number. Every figure it writes must match a value in a sealed query result, or it is withheld with its reason. How a number is proved
The platform refuses to publish a record it cannot stand behind. The refused record stays on file with its reason. A run in which two records were held
AI Native Era Business World · 7 layers
Transaction data says what happened. A decision also needs what it means and who it concerns.
Layer 1 is the semantic model agents read today. Layers 2 to 7 add meaning, ownership and context around it, in the order shown by their horizons. A layer joins only once it meets the entry rule set out under the horizons below, and no agent writes to it.
Inside the company
1
Live today
Semantic model
Warehouse and tabular models: the numbers the company trusts and agrees on.
Details
This is what agents read today, through read-only gateways. Every figure in a record is addressed to a value in a sealed query result on this model, and the record opens that evidence in one click.
2
Short term
Data ontology
What concepts mean in this company: who counts as an “active customer”.
Details
Definitions your company approves; agents read them. A record that uses a term can point to the definition it used, so two departments reading the same finding mean the same thing by it.
3
Short term
Company ontology
Units, processes, roles, and who owns which decision.
Details
Definitions your company approves; agents read them. Once ownership is known, a recommendation can be addressed to the role that owns the decision rather than to everyone who receives the brief.
KVKK/GDPR filter
4
Mid term
Behavioral data
The trace of how the company actually works: meetings, calendars, task lists and e-mail. Raw data is never taken; it passes through a KVKK/GDPR filter first.
Designed so that no record built on this layer can have a person as its subject, and every read is logged where your data protection officer can see it.
How the filter works
Direction, not shipping: none of this is in the product today.
Passes
- Nothing, until you switch a source on
- Reads for a purpose registered in advance
- Counts and durations at team level
Stops
- Any read with no registered purpose
- People, mailboxes and categories on your exclusion list
- Anything about a named person
- Message content, which stays in your environment and never reaches the model
- Groups too small to stay anonymous, which are withheld
Designed to write every read to a ledger your data protection officer and employee representatives can open. Compliance stays with you as data controller; the filter is how we limit what is read.
5
Short term
Unstructured data
Contracts, policies, procedures, decks and documents.
Details
A figure taken from a document is addressed to the document’s version and the exact passage it came from. A figure that cannot be addressed that way is not used as evidence.
6
Mid term
Tacit knowledge
Knowledge never written down, held in experience.
Details
Captured as named records: a person states what they know, the record carries their name and the date, and it is never presented as data.
Outside the company
7
Long term
External knowledge
The internet, market research and industry reports, suppliers, partners and other agents.
Details
Outside sources and other agents enter as cited, lower-trust material, read inbound only. They can add context to a record. They are never the evidence for a figure.
Accountable decision episodes
Long term
The structure that ties records to the decision moment they informed. It is not offered as a product today.
How it fits together
From data to records to knowledge
Agents read the company’s world and record every finding with its evidence. Company Context is distilled from those records, and knowledge reaches each person in the form they need. Agents read, verify and record. They never write to customer systems.
A
Enterprise systems
Structured data
Live today
- ERP
- CRM and sales
- Finance and accounting
Unstructured data
Short term
- Documents and contracts
Mid term
- Meetings and calendar
Outside world
Long term
- Internet
- Market reports
- Suppliers and partners
- Other agents
B
The world agents read
AI Native Era Business World · 7 layers
The raw world agents read, inside and outside the company.
C
AgenticObjects agents
Read · verify · record
Read-only
D
Records
Agentic objects: persistent records in which every figure has an address and evidence.
What records become
Company Context
Mid term
Company memory distilled only from published records: what the company knew, when, and on which evidence.
Actionable insights
Insight a person can check, trust and act on.
The right object for each person
The same record, in the form its reader needs.
E
Knowledge objects
Live today
- Finding What changed, with evidence
- Recommendation What can be done
- Brief A regular summary
Short term
- Stories Links findings into a narrative
- Investigation Drills deep into one question
- Perceptive Dashboard Fed by published records, as current as the latest one
Knowledge objects
Same record, a different form for each reader.
A finding, the recommendation built on it, and the brief that carries both are three record types, tied to each other by their evidence. Each keeps the link to the figures it rests on. Stories, Investigation and Perceptive Dashboard are next, built on the same records.
Live today
Short term
Live today
Illustration
Finding · CFO Agent · scheduled run
01M1B399HJ93R8075A761ESV8N
Net revenue is below its 4-week average; the gap is concentrated in two product groups.
Net revenue last month: ■■■. Its 4-week average: ■■■. Share of the gap in the two product groups: ■■■.
Evidence revenue_net@last_month · evidence 3/3 · e1.r1.c0x.w8
For the analyst and the controller: what changed, and the address every figure was read from.
Live today
Illustration
Recommendation · CFO Agent · scheduled run
Review price and discount terms in the two product groups that carry the gap.
Rests on Finding 01M1B399HJ93R8075A761ESV8N: net revenue is below its 4-week average.
Evidence revenue_net@last_month · evidence 3/3 · e1.r1.c0x.w8
For the person who owns the decision: what can be done, opened onto the finding it rests on.
Live today
Illustration
Monday Executive Briefing · CFO Agent · 08:10 · scheduled
This week: net revenue is below its 4-week average, concentrated in two product groups, and a price review is recommended.
Covers 1 finding and 1 recommendation, each opening its own record.
Evidence evidence chain open · e1.r1.c0x.w8
For the executive: the week in a few lines, with every line still one click from its evidence.
Short term
Stories
Several findings put in order as one account of what happened. Each step in the story still opens its own record and its own evidence.
Not in the product today. It is built on the same records, with the same evidence links.
Short term
Investigation
An agent takes one question further than a scheduled run would, and records what it found and what it ruled out.
Not in the product today. It is built on the same records, with the same evidence links.
Short term
Perceptive Dashboard
A dashboard whose tiles are published records, so it stays current as records are published, and every tile opens the finding and the evidence behind it.
Not in the product today. It is built on the same records, with the same evidence links.
Horizons, not dates
An order of work, not a delivery schedule
Each horizon builds on the one before it. Our rule for every new layer: it joins only when three things exist: an address for its evidence, a rule for refusing when that evidence is missing, and an access rule your administrators set.
Live today
- Layer 1: the semantic model
- AgenticObjects agents, read-only
- Records with addressed evidence
- Finding, Recommendation, Brief
Status
SHIPPING
Short term
- Layer 2: data ontology
- Layer 3: company ontology
- Layer 5: unstructured data
- Stories, Investigation, Perceptive Dashboard
Status
NEXT
Mid term
- Layer 4: behavioral data, through a KVKK/GDPR filter
- Layer 6: tacit knowledge
- Company Context
Status
LATER
Long term
- Layer 7: external knowledge, including other agents
- Accountable decision episodes: the decision moment tied to the records it rested on
Status
LATER
This page describes direction. It is not a commitment to deliver a capability on a date.
Start where it already works.
A pilot runs on what is live today: on your warehouse, with acceptance criteria written first. Everything else on this page is direction.
Agents execute their runs; they take no action in your systems. What they leave behind is the record.