Agentic systems
Custom AI that runs the repetitive work, built around how your operation actually moves.
02 · ExtensionCustom software
The applications and integrations that hold the systems together, made for your shop.
03 · SubstrateThe Polis Stack
A verified language, a supervised runtime, a reasoning engine. Written in-house, top to bottom.
What we do
An operating partner, not a vendor.
We embed in your operation, build agentic systems and custom software around how you actually work, and train every person on your team to use them. Engagements start with a consult that maps your workflows before anything is built, and everything we ship runs on a stack we built security first.
How we operate
What we hold ourselves to.
- 01
We do not state a capability before it is verified.
- 02
We do not use marketing tactics we would refuse to build into the product. No retrieval poisoning, no manufactured authority, no inflated metrics.
- 03
We say what we can prove, and we say it precisely. If a claim requires a footnote to be true, we drop the claim.
- 04
We will tell a client when AI is not the right tool, including when that answer costs a deal. A firm selling trust cannot operate by deception in any channel.
What we do · 01
Custom AI & software, built for your business, and the team that runs it.
The core of Roan Co. We embed in your operation, design agentic systems and custom software around how you actually work, and then we train every person on your team to use AI in their day-to-day.
Step 01 · Consult & design
We walk your operation and figure out where AI will actually help.
Hands-on operational walkthrough. Workflow maps for every department. A prioritized list of where AI and custom software will move the needle, and where they won't. You leave Step 1 with a roadmap you keep whether you hire us for Step 2 or not.
Step 02 · Build & train
Then we build the systems, and teach your team to use them.
Custom AI systems and software, integrated into the tools you already use. Hands-on training for every employee, from the front desk to the owner, on how to use AI in their day. We stay on for the optimization curve as your team grows into the new way of working.
Custom software
Interfaces, internal tools, dashboards, built around your actual workflow, not a template.
Agentic systems
AI agents that handle repetitive work, intake, scheduling, follow-ups, document processing.
Team training
Every employee gets walked through using AI in their role. Hands-on, not a slide deck.
Internal copilots
Private AI assistants that know your processes, your clients, and your voice.
Workflow audit
A clear map of where time is leaking, with a ranked roadmap for what to build first.
Ongoing optimization
Quarterly review, runbook updates, and on-call support as your operation evolves.
What we do · 02
Websites that match the business behind them.
Once your operation is humming, your digital presence should reflect it. Custom-built marketing sites, identity work, and the integrations that keep them in sync with the systems we just installed.
Custom marketing site
Designed and built from scratch, not a template. Fast, accessible, easy to update.
Identity & rebrand
Logo, palette, type system, and the assets your business has needed for years.
CMS & integrations
Booking, CRM, intake, wired in to the agentic systems that run behind the scenes.
Who we serve
Different industries. Same problems underneath.
If you have a team and a workflow, we have something to build. The same bottlenecks just wear different vocabulary. Here is what the work looks like in four of them.
Law firms
Intake, conflict checks, document automation, matter-based billing, follow-up loops.
Practice opsFinancial services
Client onboarding, document collection, compliance reporting, review cycles, renewal outreach.
Compliance & opsAccounting practices
Onboarding, document collection, reminders, year-end checklists, client portals.
Client opsSpecialty retail
Inventory triggers, vendor reorder, in-person and online payments, loyalty.
Floor & back-officeWhere would you start?
Pick the work that eats your week. We'll tell you what to build first.
Every Step 1 audit starts here. Check off the workflows that drain your team's time, then bring the list to a 90-minute working session.
We don't promise a percentage, we promise a roadmap. Your numbers stay yours.
What we do for banks
The version your compliance team can say yes to.
AI inside a bank has to clear a higher bar. Regulators, model risk teams, and auditors all need to know what it did and why. That is the bar we build to. We build and harden AI agent systems for regulated financial workflows: fraud and anomaly monitoring, AML and compliance workflows, document processing, and internal agents that touch sensitive data. Systems deploy inside your perimeter, on your hardware or your private cloud, with fully local model options for data that cannot leave. Agents cannot reach anything you have not allowed, a human approves anything irreversible, and every action is written to a signed, tamper-evident audit trail your risk and audit teams can verify independently. If your team wants AI and your compliance department keeps saying no, this is the version they can say yes to.
Government and defense
The highest-stakes case for security-first engineering.
Public-sector systems are the highest-stakes environment for AI, and the clearest case for security-first engineering. We build to that standard by default.
- 01
Tamper-evident audit trails for full attribution and after-action review: altering a past entry breaks verification.
- 02
Zero-trust agent runtimes: deny by default, least privilege, fail closed.
- 03
Untrusted input handled through a single, supervised boundary rather than scattered trust across the system.
- 04
On-premise deployment with fully local model options for environments where data cannot leave the building.
We work with public-sector teams on cybersecurity and AI assurance, from full builds to focused engagements. Start the conversation: calendly.com/joshualangsam/intro-call
Our technology
The Polis Stack.
We don't sit on top of someone else's AI infrastructure. We built our own, end to end. A verified language, a supervised runtime, and a reasoning engine that shows its work. Three components. One polity. One audit truth.
Verifies.
Logos
A security-typed programming language we built ourselves, designed so entire categories of security bugs cannot exist by construction: the goal is a compiler that rejects insecure code outright instead of letting it fail in production.
- An AI-native language with a syntax our team works in fluently. Pattern matching, pipes, immutable by default, strong inferred types, so generated code is correct on the first pass more often.
- Hardened by construction. The design rules entire categories of unsafe operation out at the language level instead of catching them at runtime.
- Security guarantees are structural, not bolted on. The design intent is that the dangerous path cannot be expressed in the first place.
Supervises.
Diogenes
The runtime that wraps every AI agent we ship. Acts as the watchman that catches an agent the moment it tries to do something it shouldn't, before anything reaches your systems.
- A tools layer with bounded, deny-by-default permissions. Requests an agent is not cleared to make are rejected before they leave the boundary.
- A tamper-evident audit trail. Every action lands in a record where altering any past entry breaks verification, so the log cannot be quietly rewritten.
- Untrusted input routes through a single escape boundary instead of being scattered across the codebase, so the surface that has to be trusted stays small.
Reasons.
Heraclitus
Our domain-trained AI for tax and accounting work, built so its reasoning can be checked against the rules it is supposed to follow.
- A domain-trained model for tax and accounting work, grounded in the Internal Revenue Code and the CFR rather than left to answer from memory.
- Re-derivation, not retrieval. A deterministic verifier is designed to re-compute claims from first principles: basis conservation, depreciation method, citations, entity elections, temporal validity. Claims that don't re-derive are caught before output.
- Audit-by-construction. Reasoning steps land in a tamper-evident ledger. Altering a step produces a verifiable error. The chain is designed to be demonstrable end to end.
Off-the-shelf agent frameworks fail in predictable ways. Prompt injection slips through, audit logs get edited, agents call tools they shouldn't. The Polis Stack is our answer: own the substrate, verify the language, supervise the runtime, and make claims re-derivable. It's the foundation under what we deploy.
Our mission
Security first.
Roan Co exists so that businesses never have to choose between using AI and protecting what they hold. Everything we deploy is sandboxed and supervised: agents run with bounded permissions, dangerous operations fail closed, and every action lands in a tamper-evident audit trail you can hand to a regulator, an auditor, or a client. AI you can prove, not just trust.
AI agent security
Ship agents without shipping the risk.
Most teams can demo AI. Almost nobody can put an agent in production they can actually trust. That gap is what we close. We sandbox what an agent can run, control what it can reach, and write every action to a tamper-evident record. What your agents do stays contained, attributable, and provable after the fact.
Sandboxed execution
Agents run with bounded, deny-by-default permissions. Dangerous operations route through a single boundary that fails closed: if an agent tries something it was not cleared to do, it stops.
Controlled egress
An agent cannot reach anything you have not explicitly allowed. Data does not quietly leave.
Signed audit
Every action lands in a tamper-evident audit trail. If anyone alters the record, verification fails. You can prove what your AI did, after the fact, to anyone.
How every engagement works
Consult first. Then build.
Two steps. No surprises. You see the roadmap before you see an invoice for the build.
Step 01 · Consult
We go in, map the workflows, and find the pain.
- AHands-on operational walkthrough across departments
- BPain-point map: where time and money are leaking
- CAutomation roadmap with priority & payback estimate
- DClaude Code tutoring & team training on AI tools
- EDiagnostic: what to automate, what stays human
Step 02 · Build
We design and deploy the agentic system.
- ACustom build, scoped to your specific workflows
- BIntegrated into the tools you already use
- CStaff handoff, runbooks, and on-call coverage
- DQuarterly review & ongoing optimization
- EOptional: web rebuild folded in
How we operate
Four principles we don't bend on.
The non-negotiables that shape every engagement, from first call to ongoing optimization.
01
Workflow audit before any build. We don't quote what we haven't seen.
02
Two-step engagements. Consult first, build second, each priced separately.
03
Every engagement reports to a principal. No committees, no telephone game.
04
Founder-delivered. No account managers, no offshore handoffs.
Selected work
In the field.
An early Roan Co engagement, what the audit looks like from inside an operation that's been running for thirty years.
Engagement 01 · Architectural millwork GC · Queens, NY · Step 1
Mapping a 30-year-old shop's operation, end to end.
A workflow audit across estimating, shop scheduling, change-order tracking, and AR. The roadmap surfaced a handful of agentic systems worth building in Step 2, and a few processes worth leaving human. The point of Step 1 is exactly that: knowing the difference before anyone writes code.
Step 1 consulting engagement, signed
Estimating
Quote prep & package generation
Shop ops
Job scheduling & change orders
AR & billing
Progress invoicing & follow-up
Comms
Architect & GC correspondence
The founder
One operator. Every engagement.
Every engagement is run by a principal. No account managers, no handoffs to associates you've never met.
Founder
Josh Langsam
Runs every engagement end to end: the architecture, the agent runtime, and the systems we ship. Author of open-access research on agentic AI and AI security. Leads client delivery across Florida and the Southeast.
calendly.com/joshualangsamBook a free audit
90 minutes. One operator. A real roadmap on the table.
We'll walk your operation, flag the workflows worth automating first, and outline what a Step 1 engagement would look like. No deck, no pressure, no commitment.
- ✓Hands-on walkthrough of your operation
- ✓The workflows worth automating first, ranked
- ✓What a Step 1 engagement would look like
Direct booking · replies within one business day