AI agents that take real actions — safely.
AI agent development for SaaS teams: we build agents that work inside your tools — resolving support tickets, qualifying leads, extracting data from documents — with guardrails, human approval where it matters, and cost tracking on every run.
Julian Lopez · Founder & lead AI engineer · 3 years building AI for a 1M+ learner platform
Why most of these projects stall.
Great demo, unreliable in production.
Agents loop, go off-script or fail silently on real inputs.
Actions without guardrails.
An agent with broad API access is a risk, not a feature.
Unknown cost per run.
Multi-step reasoning can burn tokens fast without tracking.
Work behind it.
Multi-agent system at a US EdTech
Julian helped build a multi-agent system across Claude, OpenAI and Gemini, connected to the platform's learning data, for a platform reaching 1M+ learners.
AI support agent
Streaming agent with tool calling (knowledge base search, ticket actions), image and PDF analysis, and per-turn cost tracking.
Agents we build
Support resolution agent
Looks up the account, checks the knowledge base and takes the action — refunds, updates, tickets — with approval rules.
Lead qualification & routing
Enriches and scores inbound leads, asks follow-up questions, and routes them to the right person.
Document processing
Turns invoices, PDFs and forms into structured data in your systems, with checks before anything is written.
Back-office operations
Automates multi-step internal workflows across your tools, with a human in the loop for exceptions.
Measured at every step. Evals, not vibes.
Map the workflow
Which decisions, which tools, and where a human must approve.
Design the tools
Typed, permission-scoped tools — the agent can only do what it should.
Build the loop
Stateful agents (LangGraph) with step limits, retries and human-in-the-loop.
Evaluate & trace
Scenario evals, full traces of every run, and cost per task.
Priced up front, in writing.
AI agents
- Fixed scope, timeline and price before you commit
- Milestones with written acceptance criteria
- Code, docs and quality tests are yours
AI Production Audit
Already have a prototype? In five business days we measure quality, cost and latency, and hand you a prioritized, fixed-price plan.
See what's in the audit →Introductory founding-client rates, for up to three portfolio partnerships — in exchange for a case study if the work produces useful results.
Send a project brief
Tell us which workflow you want automated, which tools it touches, and where a human must approve. We reply within one business day.
Questions, answered.
Which frameworks do you use?
LangGraph for stateful agents, the Vercel AI SDK or provider SDKs for simpler loops, and Langfuse for tracing. We pick the lightest thing that works.
What if the agent makes a mistake?
Risky actions require approval, tools are scoped to least privilege, and every run is traced so mistakes are visible and fixable.
Can it connect to our tools?
If it has an API, yes — CRMs, help desks, databases, internal services. We can also expose your systems as MCP tools.
How do we know it's working?
We agree on scenario-based evals up front (for example, "resolves 8 of 10 refund requests correctly") and report against them.
Tell us what you're building.
Book a free 30-minute call. You'll get a straight answer — and, if it's a fit, a concrete fixed-price plan. Sometimes the answer is "you don't need us."