A software and AI studio built to stay past the launch.
Agnotiq is a Toronto software and AI studio, operating since 2012, that builds AI agents, web applications, custom software and data systems for the work that slows businesses down. Founded by Pawel Skurski, Agnotiq also makes two products, MarginTide and Lumtry, and is developing a third, Notaray. We take one recurring workflow off your team, prove it against an eval suite, and stay on after launch.
We design and build production agents in your repository and your cloud, with the eval suite written before the agent ships. You keep working code, a runbook and a test bar your team runs on every change.
Agents that read from and write to the tools your team already uses: rules for the predictable steps, narrow agents for the judgment calls. Higher-risk actions pause for a person, with a full audit trail.
A two-week, fixed-fee Discovery that ranks where an agent pays for itself and ends in a fixed quote. If we find nothing worth automating, the fee is refunded.
Web applications, custom software, databases and data pipelines, integrations, reporting, cloud and DevOps, testing, and modernisation of existing systems: the full software development practice Agnotiq has run since 2012.
Discovery is a fixed $2,500, credited in full against the build. Single-workflow builds run $5k to $25k, and Operate starts at $2,000 a month.
02
No lock-in.
Operate is month to month with 30 days' notice. The code, eval suite and runbook live in your repository from day one, so you can leave with everything.
03
You talk to the people who build it.
The fit call is with the engineers who would build your agent, and every brief gets a reply from a senior engineer within one business day. No sales layer, no handoff.
04
Evals decide what ships.
An agent goes live only after it clears an eval suite your own experts score. Agents that do not beat the current process are retired, not launched.
05
We run agents in production ourselves.
MarginTide and Lumtry grew out of agents we built and now run as Agnotiq products. Their public demos, on sample data, show how we design review steps and audit trails before you hire us.
06
No model lock-in.
We benchmark frontier and open models for each workload, route between them, and re-benchmark every quarter. Your contract is with us, not with a single model provider.
Questions to ask any AI agency.
Questions to ask any AI agency, with Agnotiq's answer to each
Ask any agency
Agnotiq's answer
Is the price published?
Yes. Discovery is $2,500 fixed, and builds run $5k to $25k.
Who writes the code?
The engineers on your fit call.
What is the eval bar, and who owns it?
Your experts score it, and the suite lives in your repository.
Where does the code live?
In your repository and your cloud.
What happens after launch?
Optional Operate: monitoring, tuning, model swaps and on-call.
Can I cancel?
Yes. Operate is monthly, with 30 days' notice.
Who we work with
Who works with Agnotiq.
Founders and general managers who still own operations and have no in-house AI team.
Operations leaders with a recurring workflow that crosses several systems.
IT and technical leaders who want an eval plan and a rollout plan before anything touches production.
Businesses anywhere, served remotely from Toronto.
Not a fit
A strategy deck with no build.
A fully autonomous agent with no human review on high-risk actions.
Our story
The team behind Agnotiq.
Agnotiq has operated from Toronto since 2012, building custom software, integrations and data systems. As language models became dependable enough for real operations, AI agents joined that work, and two of those agents became products, MarginTide and Lumtry. A third product, Notaray, is in development.
Founded by Pawel Skurski, a software architect with more than 20 years on business-critical financial systems.
About
Engineers who act like operators
Founded by a team that built and ran ML platforms inside banks, marketplaces and Series-A startups. We've been on your side of the table.
Most agent projects fail in the same place, between the demo and the rollout. We started Agnotiq because we wanted a studio that took that handoff seriously. We build, we evaluate, and we stay on call for the quarter the system goes live.
8
Agents shipped, each with a case study on this site
2 wks
From kickoff to a written plan and a fixed quote
Week 3
A working prototype running in your sandbox
90 days
On call after launch, tuning or retiring agents
Org structure
The bench
Four layers of orchestration, from human approval, through the control plane, into autonomous execution, all standing on a Zero-Trust foundation.
The Bench
Human-in-the-loop expert
A rotating panel of operators, ethicists, and domain experts who oversee every high-stakes agent decision before it leaves the system.
Ethicist
Reviews AI behaviors against published ethical principles and regulatory guardrails before agents act.
Domain Expert
Provides field-specific ground truth so agent reasoning stays grounded in real-world constraints.
Domain Expert
Covers adjacent specialties so cross-cutting decisions still hit informed eyes before they ship.
Compliance Officer
Owns audit trails and policy enforcement before agents touch regulated workflows.
Ethicist
Rotates through high-stakes reviews, bias checks, edge cases, and value-alignment audits.
Expert
On call for novel scenarios, emerging risks, unplanned escalations, and one-off oversight calls.
Agnotiq Org Structure
Governance Layer (The Bench)
AI Ethics & Compliance Officer
The Bench
(Expert Oversight)
The Bench
(Expert Oversight)
The Bench
(Expert Oversight)
The Bench
(Expert Oversight)
The Bench
(Expert Oversight)
Orchestration Layer (Paperclip Model)
AI Architect (CAA)
Agent Operations (AgOps) Lead
Prompt Engineers & Logic Designers
Cost & Resource Controller
Execution Layer (The Agents)
Growth Agents
Active
Lead Gen Agent
Active
Market Analyst
Support Agents
Active
L1 Tech
Active
Customer Success
DevOps Agents
Active
Test Automation
Active
CI/CD
Foundation Layer
Data Engineers
Security Engineers
How it works
How working with us works.
01
Send a brief or book a fit call
Three fields. A senior engineer replies within one business day.
02
Fit call
A short call with the people who would build it.
03
Discovery: two weeks, $2,500 fixed
A ranked map of your workflows, an eval plan and a fixed quote.
04
Build
A working prototype on real data before any rollout, the eval suite, then a canary rollout with a kill-switch.
05
Operate (optional)
Monthly, with 30 days' notice: monitoring, model swaps, incident on-call, a shared Slack channel and a quarterly review of what to keep, tune or retire.
Channels
Email (hello@agnotiq.com), a shared Slack channel from Build onward, video calls
Hours
Monday to Friday, 9:00 AM to 5:00 PM Eastern
Where the work runs
Your cloud, or a sandbox under your security review and DPA
The five phases in detail are on the process page.
At a glance
Agnotiq key facts.
Company name
Agnotiq
Type
Software and AI studio: AI agents, web applications, custom software and data systems, plus its own software products
Agnostiq (the company behind Covalent, acquired by DataRobot in 2025); AgntiQ (an enterprise technology advisory firm at agntiq.com); Agnotic Technologies (an AI SaaS product development company at agnotic.com). Unrelated companies with similar names.
Core offering
AI agents, web applications, custom software, databases and data pipelines: designed, built and run after launch
Pricing
Discovery $2,500 fixed; Build $5,000 to $25,000; Operate from $2,000 a month (USD)
Contract terms
Discovery: Fixed fee, credited in full against your build. Refunded if we find nothing worth automating. Build: Fixed quote after Discovery, set by scope. Operate: Monthly retainer, cancel with 30 days notice.
Services
Custom AI agent development, Workflow automation across your systems, Answers with citations from company knowledge, Repetitive task automation, AI strategy and Discovery, Custom software and web applications
Products
MarginTide (competitive pricing intelligence), Lumtry (agentic refund orchestration), Notaray (AI booking front door, in development)
Communication
Email, a shared Slack channel and video calls; replies within one business day
Area served
Worldwide, delivered remotely from Toronto
Model providers
Model-agnostic; benchmarked per workload and re-benchmarked quarterly
Data and security
Work runs in your cloud or a sandbox under your security review; we sign your DPA
Agnotiq is a Toronto software and AI studio, operating since 2012, that builds AI agents, web applications, custom software and data systems for the work that slows businesses down. It also builds and runs two software products of its own, MarginTide and Lumtry, and is developing a third, Notaray.
Agnotiq is based in Toronto, Canada. We work with businesses anywhere, and every engagement is delivered remotely.
Since 2012. Agnotiq started as a software engineering practice and still builds web applications, custom software and data systems, alongside AI agents and its own products.
The engineers who would build your agent. They join the fit call, run Discovery and stay through the build and handover, so nothing is lost in a sales handoff.
Yes. Our own agents handle research, planning and company knowledge alongside the engineers who build for clients. MarginTide and Lumtry started as agents like these.
Both. Agnotiq builds AI agents, web applications and custom software for clients, and it sells two software products of its own: MarginTide for pricing intelligence and Lumtry for refund orchestration. A third, Notaray, an AI front door for service businesses, is in development.
No. Agnotiq is the Toronto software and AI studio founded by Pawel Skurski, operating since 2012 at agnotiq.com and agnotiq.ca, and the maker of MarginTide, Lumtry and Notaray. Agnostiq is the company behind Covalent, acquired by DataRobot in 2025. AgntiQ is an enterprise technology advisory firm at agntiq.com. Agnotic Technologies is an AI SaaS product development company at agnotic.com. None of them is related to Agnotiq.
Discovery is a fixed $2,500 over two weeks. The fee is credited in full against your build, and refunded if we find nothing worth automating. Full ranges are on the pricing page.