Agents & Automations # Give your team back their best hours. FM designs and builds the systems that take routine operator work off your team — copying data, building reports, routing approvals, synthesis. AI agents where the work needs reasoning. Plain automation where it doesn’t. TL;DR Who it’s for Small and mid-sized companies whose teams are buried in routine operational work. What FM delivers Agents and automations built on the right stack for each job — not locked into one. Engagement length 4–12 weeks per build. ## What is Agents & Automations? Every business runs on two kinds of work. The work of knowing — collecting information, watching for patterns, assembling reports, routing approvals — used to require a person. The work of deciding — judgment, taste, accountability — still does. Agents and automations move the first kind off your team so they can focus on the second. The distinction between an agent and an automation is practical, not philosophical. Automations follow rules: when X happens, do Y. Agents reason: figure out the right Y based on context. FM uses both, picks the right one for each job, and combines them where it makes sense. Most real systems are a mix. Done well, the result isn’t a chatbot bolted onto your business. It’s the boring parts of your operators’ day going away — and the work that’s left being more interesting, more important, and more leveraged. Click to expand Watch FM co-founder Tim Visconti on how to get started with automations. Tap the video to play in full screen. ## When do you need Agents & Automations? - Your team copies data between systems for hours every week. - Reports that should take an hour take three days, every month. - Approval workflows depend on someone remembering to ping someone. - Lead routing, CRM hygiene, or pipeline reporting eats your operators’ time. - You have a clear, repeated process that someone keeps having to babysit. - You want to deploy agents that reach into your business systems, but don’t know where to start. ## When you don’t. - The work changes daily and requires judgment in every case. Automating it will create more problems than it solves. - You don’t yet have a stable process to automate. Start with AI Adoption to find the right targets first. - You want a low-code DIY effort. FM can help you scope one, but the right partner for that is your own ops team plus a Zapier seat. - The work happens fewer than a dozen times a year. The build cost won’t pay back. ## How does FM approach Agents & Automations? Agents and Automations is engineering work, so the tools matter. FM picks the right stack for each job — visual automation where non-engineers will extend it, agent frameworks where reasoning is needed, custom code where neither fits. ### Tools and frameworks FM uses #### n8n Visual workflow automation. Used when the rules are clear and the system needs to be readable and extendable by non-engineers on your team. #### MCP (Model Context Protocol) The connective tissue between AI models and your business systems. FM builds custom MCP servers so Claude, ChatGPT, or your own agents can reach the data they need without one-off integrations. #### OpenClaw / Hermes Custom agent frameworks. OpenClaw handles complex, multi-step work where an agent needs to coordinate tools, manage state, and recover from failure. Hermes handles narrower, scoped deployments where OpenClaw would be overkill. #### Claude (and other frontier models) The reasoning layer. FM evaluates which model fits each use case rather than defaulting to a single provider. #### Evals and observability Custom evaluation suites and structured logging built into every system FM ships. The difference between a demo and a production system is knowing when it breaks before your users do. #### Custom code For the cases where no framework is the right fit. FM writes production code in TypeScript and Python rather than forcing every problem into a tool that wasn’t built for it. ### How an engagement runs Phase 1 · 1–2 weeks #### Discovery - Workflow mapped end-to-end, including the unwritten steps people do from memory - Decision on tool stack: visual automation, agent framework, custom code, or a mix - Risk and failure-mode analysis: what happens when the system gets it wrong - Acceptance criteria written in plain language, not technical specs Phase 2 · 2–6 weeks #### Build - Working system, deployed in your environment, exercised against real data - Human-in-the-loop checkpoints where the work needs judgment - Monitoring and logging your team can actually read - Iteration cycles based on what shows up under real use, not just demo data Phase 3 · 1–2 weeks #### Handoff - Documentation that lets your team operate and extend the system without FM - Training sessions with the operators who will use it day-to-day - A maintenance plan: what FM owns ongoing vs. what your team owns - A retrospective on what worked, what didn’t, and what to build next Specific shapes this work takes ### n8n Implementations Visual workflow automation set up the right way, so non-engineers can extend it without breaking what already works. ### Custom Agent Systems Custom agentic deployments and automations built on OpenClaw, Hermes, and other modern agent frameworks — the right tool for the job rather than locked into one. ### MCP Servers MCP servers built around your data and workflows. Connect Claude, ChatGPT, and your own agents to the systems they actually need — without one-off integrations for every tool. ### AI Agents for Sales Ops Lead routing, CRM hygiene, and pipeline reporting handled by agents instead of analysts. ### Customer Service Enablers Agents and copilots that handle the routine and route the rest to your team with context already loaded. ### Content Operations AI-powered content production that scales without adding headcount. Plan, draft, and ship without burning out your writers. [ ![FM’s AI Persona Tool — turning scattered GPT experiments into a proprietary AI asset.](/_next/image?url=%2Fimages%2Fposts%2Fai-persona-tool-case-study.jpg&w=3840&q=75&dpl=dpl_EdcU1uSCP5BSti6JnVxZvoni8wGn) Featured case study ### AI Persona Tool — From Generic GPT to Proprietary AI Asset How FM built a custom AI persona system that turned a client’s scattered GPT experiments into a proprietary, reusable asset their team could actually rely on. Read the case study ](/resource-center/ai-persona-tool-case-study) ## Frequently asked questions. ### What’s the difference between an agent and an automation, and how do you decide which to use? Automations follow rules; agents reason. If the work is "when X, do Y" with clear inputs and outputs, automation is faster, cheaper, and more reliable. If the work needs the system to figure out the right next step — pulling context from multiple places, handling exceptions, or generating language — an agent earns its place. Most real systems combine both. FM picks the simpler tool first and adds reasoning only where it pays back. ### How much does an agents and automations engagement cost? It depends on scope, the systems involved, and how much custom code is required. FM shares a specific range after a 30-minute scoping call. As a frame of reference: most engagements run 4–12 weeks of focused build time and are priced on outcomes rather than hours. ### Will we own the code and the system after the engagement? Yes. You own everything FM builds — the code, the configurations, the documentation. There is no licensing, no per-seat fee, no platform you have to keep paying FM to access. If you want FM to keep maintaining the system, that’s a separate retainer; if you want to take it in-house, the handoff is built into the engagement. ### What if our team doesn’t know n8n, MCP, or agent frameworks? Most clients don’t. The handoff phase exists for exactly this reason — FM documents what was built, trains your operators, and structures the system so non-engineers on your team can extend it. Where deeper technical work is needed later, you can come back to FM, hire someone, or use the documentation to evaluate other partners. ### How do you handle the AI doing something wrong — hallucinations, bad data, or unexpected behavior? Three ways. First, FM builds human-in-the-loop checkpoints wherever the cost of being wrong is meaningful. Second, the systems log their reasoning so when something goes sideways you can see why. Third, FM doesn’t pretend the AI is reliable on tasks where it isn’t — automation handles the deterministic parts; reasoning is reserved for places where it genuinely earns its place. ### Do you support the systems you build after launch? Yes, on a retainer basis. Many clients keep FM engaged for ongoing monitoring, optimization, and new feature work. Others take the system in-house once the team is up to speed. Either way is fine — the handoff is structured so you have a real choice. ## Ready to talk about Agents & Automations? Reach out and we’ll set up a conversation. No hard sell, no decks — just a working session to see whether FM is the right fit. [Let’s Talk](/get-started)[See all solutions](/solutions) ### Related pillars - [AI Adoption](/solutions/ai-adoption) - [Custom Software](/solutions/custom-software)