Expogentic, by Puente Global

The digital workspace 2.0.

A company of ten, with the output of a hundred.

Your people and your agents, working from one record. Keep the tools you like. Replace the ones you do not, and stop paying for them.

A small stone workshop at night, every window lit and full of people at work.
Overnight, an example run4 agents, 1 workspace
02:14OperationsFiled the morning brief from yesterday's calls, messages and board. Two decisions are waiting for you.done
03:40KnowledgeFiled the site walkthrough as a page, transcript attached. Three things agreed on it are now dated tasks.done
05:02SalesDrafted the reply to a quote request from the website. Waiting for you to send it.your call
06:15DeliveryProposed an estimate and a next step for the nine cards that had none. Waiting in Review.your call

An example night, not a customer's figures. Nothing above was sent, paid or published.

Why most of this fails

MIT NANDA, The GenAI Divide: State of AI in Business, 2025

Only 5% of AI investment ever reaches the P&L.

In MIT's 2025 study, 95% of company AI pilots produced no measurable effect on profit and loss. Most stalled between the demo and the day job.

The 5% that landed share one trait: memory and learning inside a real workflow. Expogentic is built around that trait, and the rest of this page shows it.

Retain

Living Pages

Versioned and zero-delete, so knowledge compounds instead of evaporating in a chat window.

Context

Knowledge Base

Agents read your systems directly. Schema gated, role gated, logged.

Improve

Persistent memory

People, projects and promises, carried across every run. On day ninety it still knows day one.

Every time the models get better, your agents get better, on the same knowledge. The return compounds.

A telephone switchboard, every cord running into one socket.

The question everyone asks

So why not Jira, Slack, Notion and Airtable, with AI switched on?

Four good tools, each built for people, with the AI added beside them. The difference is structural.

Four tools with AI addedExpogentic
Who is the agent?A feature inside your seat. It acts as you.A colleague with its own account, its own permission group and its own schedule.
What can it see?The page you have open, and whatever you paste.Your company's whole Knowledge Base, fenced by the same grants your people get.
Where does the data live?Their database, shared with every other customer.Your own database, one per company, read live.
When does it work?While you are typing at it.On a schedule, overnight, and it reports back into the chat.
What does it replace?Nothing. It sits beside all four.Keep what you like and connect it. Replace what you do not, and stop paying for it.

Inside the workspace

Six rooms of one workspace, and what an agent does in each.

08:04

The day is already written.

Overnight your agent read the calls, the messages and the board, and filed the brief before the office opened.

09:15

It proposes. You accept.

Every agent edit, row change, task and lead waits in one queue, kept exactly as proposed. A person accepts it, or does not.

11:30

The board is the database.

Drag a card and the row moves. There is no second copy of the pipeline to reconcile on Monday.

13:00

One thread, both kinds of colleague.

People and agents in the same conversation, answering from the same record. Ask the same thing in your own Claude over MCP and get the same answer, capped at what you may see.

16:00

The call files itself.

Every recorded meeting arrives as a page with its transcript, and what was agreed becomes dated work on the board.

17:12

It drafts. You send.

Outbound waits in the approval queue, addressed only to somebody on your trusted list. A person presses send.

Brief

Prepared for you overnight. It stays until you say you are caught up.

Waiting on a person

  • Phase two scope: fixed price or day rate.
  • The install slips a week unless the rail order goes today.

Commitments, now dated tasks

  • Send the revised phase two scopedue Thu
  • Confirm the delivery windowdue Tue
  • Chase the rail orderdue today

At risk (2)

TaskOwnerWhy
Chase the rail orderRosaDue today, no update since Monday
Phase two quoteTeodorWaiting on the client for six days

Review

Changes somebody has proposed but cannot apply themselves. Whatever you decide is recorded against your name.

PagesInstall standards: exclusions list, two clauses addedfrom the operations agentChanged since proposed
Tablesdeals: Mezzanine install, stage set to wonfrom the operations agent
ColumnProposed value
stagewon
CRMNew lead from the website: a fit out in Leeds, budget statedfrom the website agent

Pipeline

12 rows, lanes by stage, drag a card to change itOpen table

Proposal sent 2

Survey, Harrogate

18,400Rosa12 Sep

Fit out, Leeds

31,500Teodor

In progress 1

Maintenance, Q4

9,200Rosa30 SepRenewal

Won 1

Mezzanine install

46,000Teodor

deals.stage set to won, by the person who dragged it

Chat

Rosa, Teodor and the operations agent

Rosa9:41

Can the install date hold?

Operations agentAI9:41

Yes, if the rail ships Tuesday. I have dated the chase for today and put it on the board.

Teodor9:52

Fine by me. Send the client the revised dates.

Operations agentAI9:53

Drafted. Waiting in the approval queue.

My Meetings

Recorded 12 meetings

Site walkthroughToday, 42 minWith: Rosa, Teodor3 to do
Weekly deliveryYesterday, 28 minNothing outstanding

The record

  • Revised scope agreed for Thursday.
  • Rail order confirmed for Tuesday delivery.

Agreed: revised scope by Thursday.

became

  • Send the revised phase two scopedue Thu

Full transcript

Agents

Each agent carries its own permission group, decided the same way it is for a person.

AgentDoesPermission group
OOperations agent ReadyOperations DirectorOperations agentConfigure agent
WWebsite agent ReadyMarketingWebsite agentConfigure agent

Approvals

Toteodor@example.comOn the trusted list

SubjectRevised phase two scope

The revised scope is attached. The dates hold.

Drafted 17:12 by the operations agent

What a colleague means, exactly

An agent has an account, not an API key.

The matrix that decides what a person may read decides what an agent may read. You can see every grant, and change any of them.

Permission groupTasksPeople & CRMPagesChatOutbound emailagentsTables
LeadershipFullFullFullFullFullFull
StaffEditView & suggestEditEditNoneView & suggest
Operations agentAIEditView & suggestEditEditHeld for approvalView & suggest
Website agentAINoneNoneView & suggestEditNoneNone

None means the agent does not know the table exists.

A staff member, in their own ClaudeHow much did we sell last month?
The agentFinance is not in your permission group, so I cannot show revenue. Here is what you may see: closed jobs by service.

The wall that tells a person no tells the agent no too.

A stone office building at night across a canal, three windows lit, each a different room.

Business hours end. The work does not.

Expogentic runs the night shift.

The cast

Five agents. Each one hired for a job.

Every workspace names its own. These five are examples, drawn from what agents do in ours.

  • Odile, drawn as a grey heron in a charcoal wool jacket against a flat deep green field

    Odile

    Operations Director

    • Files the morning brief before anyone logs in.
    • Turns what was agreed on a call into dated tasks.
    • Notices when work has slowed and books the meeting.

    tasks: editcrm: viewemail: held

  • Bram, drawn as a red fox in a navy blazer against a flat warm ochre field

    Bram

    Sales and outreach

    • Answers visitors on your website and captures the lead.
    • Drafts the reply to a quote request, held for a person to send.
    • Sends the campaign, reads the replies, learns which message worked.

    crm: editpages: viewemail: held

  • Tamsin, drawn as a tawny owl in a dark green cardigan with round spectacles against a flat stone field

    Tamsin

    Finance

    • Reconciles what was invoiced against what was earned.
    • Rebuilds the forecast from the pipeline, not from memory.
    • Chases what is owed and flags a renewal with no owner.

    finance: editcrm: viewsend: never

  • Ferro, drawn as a black ox in a rust work jacket against a flat dusk blue field

    Ferro

    Delivery

    • Keeps every card to one standard, so estimates become projections.
    • Reports the week into the channel where the team already is.
    • Asks before it changes a date.

    tasks: editchat: editemail: none

  • Ines, drawn as an elephant in a soft grey suit jacket against a flat brick red field

    Ines

    Knowledge and meetings

    • Files every recorded meeting as a page, transcript attached.
    • Answers what was decided with a client in May, and cites the meeting.
    • Answers in your own Claude over MCP, capped at your permissions.

    pages: edittables: viewemail: none

What an agent is

Role-shaped, not role-replacing.

People were made for creative work, and nobody should spend their best hours on work a machine should have carried.

It does the work
The board, the books, the outreach. The function, not a summary of it.
It brings analysis
What is slipping, what the numbers mean, what to do next.
It never claims the seat
You stay accountable. Nothing leaves without a person.
Two people sketching at a drafting table while a heron in a jacket carries folders to a filing cabinet.

The shape of it

The same company, with a larger surface.

A long limestone building at dawn, the old half and the new half lit identically, the last bay in scaffolding.

What stops happening

  • The status meeting that only moves information between two people.
  • The Monday reconstruction of where everything stands.
  • The CRM three weeks stale because updating it is nobody's job.
  • Context walking out of the building when a person does.

What starts

  • The brief is filed before you arrive.
  • The CRM updates itself from the call.
  • The board and the database are one thing.
  • What a person knew stays when they leave.

Where your data lives

A green steel strongroom door standing open, a filing cabinet on wheels inside, ready to roll out.

Your data, in your own database.

Does my data train somebody's model?

No. Your data is context for a call, never training for a model.

Where does it sit?

Each company's tables live in that company's own database, and its files in its own store. Every answer is capped by the least privileged person who will read it.

Can I leave?

The database goes with you. It is ordinary Postgres, readable with ordinary SQL.

Answers

Questions you might have, if you are discovering Exponential Agentic Growth.

How is this different from the AI my team already uses?

That one waits to be prompted. These hold an account, a permission group and a schedule, and work while nobody is watching.

Can an agent email somebody outside the company?

Only through the approval queue, to an address on your trusted list. A person presses send.

Does this replace people?

No. It carries the chasing, filing, drafting and remembering. Every example on this page ends with a person deciding.

Where do I start?

Ask below. We onboard a small number of companies by hand, so the Knowledge Base is set up properly rather than left empty.

Which tools does it work in today?

Email both ways, Slack both ways, and MCP with a token per person, so your own Claude reaches the Knowledge Base. Jira, Git and GitHub are coming.

Ask for a workspace.

Tell us who you are and what your team spends the week on. A person reads every request.

This form creates no account. Access is opened by a person, one workspace at a time.

Want to sell this to your own clients? Say so in the optional field above.