AI engineers,inside your operation.
Forward-deployed engineers who build, run and hand over AI agents that live in production.
Built by the team behind
PodsnagMost companies your size already tried generative AI. ChatGPT Enterprise, an AI agency, an internal pilot. Almost nobody got an agent into production that moves the number.
The barrier is not the model. It is wiring it into the operation.
of corporate generative AI pilots deliver no measurable impact on the P&L.
Source · MIT NANDA, “The GenAI Divide: State of AI in Business 2025”
The pilot
It looked good in the deck and went back in the drawer.
The agency
Slides, not operations. The contract ends, the project ends.
The hire
Six months to hire, twice the salary in payroll charges, and the best candidate goes to a global startup.
It takes someone writing the code that makes it part of the operation.
We wire AI into the stack you already run.
---------+ +--
| |
------------+ +--
+-------- --------+
---------+ +--
| |
-----------+ +-- WHATSAPP TOTVS ERP
GMAIL +======================+ HUBSPOT
| MARFIN AGENT |
NF-E XML +======================+ SLACK
SHEETS POSTGRES > order #4821 · whatsapp -> totvs erp · created in 1.2s_
Agents that do the work. Inside the systems your team already uses.
Answers every WhatsApp order
Distribuidora Sul
40 boxes for Thursday?✓ You: Confirmed: order #4821, 40 boxes, delivery Thursday before 2pm.
Mercado Bom Preço
Can you resend the invoice?
Rede Farma
Thanks, received.
Hotel Atlântico
Price for 200 units?
Distribuidora Sul is waiting
Asked for 40 boxes by Thursday. Stock checked in the ERP.
Reconciles every invoice against the ERP
Reads from
143 NF-e XMLinbox
Purchase ordersERP
Approval limitssheet
Finance agent APP
NF-e 35.244 from Química Vale is R$ 1.240 above purchase order 7781. Approve the difference?
Updates the CRM after every sales call
Call · Rede Farma
23 min · transcribed
Ana: we need it live before the Q3 audit.
Ana: budget is around R$ 48 thousand, send me the quote by Friday.
Rede Farma · Deal
Reports the numbers every Monday at 8am
Weekly operations report. Three things need you:
OTIF
96.4% ▲ 1.2
Backlog
112 ▼ 18
Gross margin
18.2% ▲ 0.4
- · Supplier Química Vale late twice this week
- · Route SP-04 cost up 11% since fuel change
- · SKU 4471 runs out in 6 days at current pace
From first call to production in 16 weeks. Engineers who work from inside your operation.
01 · Weeks 1–2
Discovery
Scope document, POC plan and go / no-go criteria.
scope · order-intake.md
02 · Weeks 3–6
POC
A working agent in staging and a report with the production decision.
eval · 1,200 real orders
96% accurate
Decision: go to production
03 · Weeks 7–14
Production
Agent live on your systems, with evals, dashboards and the first handover.
Live · order-intake-agent
p95 1.2s · R$0.04 / run
12,480 orders handled this week
04 · Month 4 onward
Operation
Embedded squad, new use cases, progressive handover to your team.
next quarter
Marfin Agents comes pre-installed. The platform behind our own products. First agent in weeks, not months.
Orchestration
Multi-step runs across your systems, with retries and a human in the loop where it matters.
incoming · finance-agent
Refund R$ 7.200 to João Silva, CPF 123.456.789-00, order 4821.
Guardrails
Personal data masked before the model sees it. Policies that stop money from moving without a person.
p95 1.2s · R$0.04/run
Evals and observability
Every version measured against real cases. The layer that separates a toy agent from one that survives an audit.
1k+ connectors
ERP, WhatsApp, CRM, email, spreadsheets, databases. The stack you already run, wired in.
You hire the team that built these.
Five AI products in production since 2023. Each one with a live platform, active users and an in-house team.
active users across our products
agents in production
active integrations
Three entry points: discovery, fixed-scope project, embedded squad. You leave the first call with the price range settled, before any proposal scoping.
01 · Discovery + POC
We build a working pilot in 4 to 6 weeks.
02 · Fixed-scope project
Agent in production in 8 to 16 weeks.
We map a critical process, build a working pilot, and show the production plan with pricing. No lock-in.
- 4 to 6 weeks
- Process mapping
- Working POC in staging
- Production plan with pricing
Integration with your systems, continuous evals, LGPD governance, training for your team.
- 8 to 16 weeks
- Integration with legacy systems
- Continuous evals and guardrails
- In-house team training
03 · Embedded squad
1 to 3 engineers inside your operation.
Rolling roadmap of new use cases, continuous improvement of the agents already running, access to the Marfin Agents platform and ongoing support. 3-month minimum contract. Starting at R$ 35k per month, per engineer.
Questions.
Anything else, ask on the call. Thirty minutes, and if it does not make sense, no friction.
Book a discovery callWhat is a Forward-Deployed Engineer?
An engineer who works from inside the client's operation instead of from a vendor's office. The term comes from Palantir; OpenAI and Anthropic run the same model today. The FDE sits with your team, ships against your real systems and data, and stays until the agent runs in production and your people can operate it.
How is this different from a consultancy like Deloitte or EY?
Consultancies sell a diagnosis and a deck. We sell running code. We measure success by agents in production and impact on your business, not by billable hours or status reports.
Are you a software house?
No. A software house rents you devs by the hour, with no opinion on AI architecture. We have our own platform (Marfin Agents), we have run AI products for years, and this team has deep command of LLMs, evals and MLOps.
How long until I have an agent in production?
Between 8 and 16 weeks for the first use case, depending on integration complexity. Discovery + POC is ready in 4 to 6 weeks.
Who owns the code, us or you?
The agent code is yours. The Marfin Agents platform is licensed (or run in your own environment, if you prefer). You are never locked in with us.
What about LGPD?
Every engagement includes data flow analysis, consent modeling and audit documentation. We work with your legal team from Discovery onward.
How does it start?
You book 30 minutes with us. We learn the process, check whether it makes sense, and propose a Discovery. If it does not make sense, no friction.
Which sectors do you work in?
Today's priorities: manufacturing, agribusiness, retail, healthcare, mid-market finance, edtech and logistics. Outside that, case by case.
What if I already have an in-house team?
Even better. Our model includes capability transfer. We measure it every quarter. The goal is for your team to become autonomous, with us as the scaling partner.
Discovery call · 30 minutes
Book the first call.
Tell us the process. We check whether it makes sense and propose a Discovery. If it does not, no friction.
Agents in production. Not in the drawer.