The story behind this case
Instead of opening requests on the platform and waiting for permission, teams were given an isolated environment — to reduce the impact of untrusted or experimental code on shared clusters at large scale.
Crash Course: Always-on agents
Mentor: Denys Vasyliev, Principal Site Reliability Engineer / UK Global Talent Visa Holder, 17+ years in the industry.
Crash Course: Always-on agents
about the course
An always-on agent keeps working when nobody is asking. It watches a system, remembers what it has seen, decides what matters, acts within limits and checks its own work. Six sessions take those parts in turn, and each has you run one.
The course follows the architecture of Triage Agent: from the agent at the edge of the system that collects signals, through the core with a single memory, xray, Tria and the spot, to ego, the recursive self-improvement system that scores agent sessions from their traces. Across the six sessions you run the full loop once: signal in, decision made, memory written, a teammate asked, a task handed to an agent, work scored.
Each session lasts 2 hours: a talk, a lab and a discussion.
Mentor: Denys Vasyliev, Principal Site Reliability Engineer.
You’ll work through all of this using triageagent.dev — an open example of a developer environment built on agent substrate. Triageagent is the winner of the National Award India AI Gamechanger 2026.
What you take away
- A map of the always-on loop (sense, decide, remember, act, evaluate) and where each Triage component sits on it
- The ability to read a service through USE, RED and the four golden signals, and to say which family fits which question
- A working picture of System 1 and System 2 in an operations agent: why most events end at a reflex rule or a small model, and what makes a slow loop worth running
- Hands-on use of agent memory over MCP: remember, recall and forget from any agent, with search by meaning and a linked graph
- Practice handing a task to an always-on agent and following it through its harness, sandbox, snapshot and resume
- A repeatable way to evaluate agents from traces they already emit, and to turn a score drop into one targeted change
format
Duration6 sessions of ~2 hours each
StartNovember 30 at 6:30 PM (Kyiv time), with sessions on Mondays, Wednesdays, and Fridays
PlatformZoom, the link will be sent the day before the course
Course languageUkrainian
Slides languageEnglish
SeatsLimited
LevelWorking knowledge of Kubernetes
audience
- SRE engineers
- DevOps engineers
- Platform engineers
- Backend engineers
program
Six sessions: from collecting signals to evaluating agents
Each session: talk 40 min, lab 60 min, discussion 20 min
NameTopicSession
session-1 The agent and its connectorsSENSE · 120 min 1/6 ›
What you’ll learn
- – What always-on means: a loop that never stops, and the five verbs of the course (sense, decide, remember, act, evaluate)
- – A first look at the live demo on desktop, phone and watch
- – The collection layer: it sits where your systems run and sends signals only; issues and root causes are decided in the core
- – One agent covers Kubernetes clusters and Linux hosts, with connectors to cloud providers
- – It discovers virtual apps from your workloads and measures them with eBPF, cgroups and logs-based metrics as USE, RED, SIG and LOG signals
- – Everything is sent over one outbound secure WebSocket
Lab 1 · SENSE
Install the agent against a sample workload, watch auto discovery build the service map, and read the same service through USE, RED and SIG.
session-2 The core and its two systemsDECIDE · 120 min 2/6 ›
What you’ll learn
- – Each event is sorted on arrival as noise, signal or issue, by reflex rules and an embedding model
- – Most events never reach a large model
- – Jev decides what passes the guardrail, where an event is routed, how severe it is and what is likely to follow
- – Slower loops reflect on resolved issues, reason about causes and write what they learn back to memory as facts and patterns
Lab 2 · DECIDE
Emit a mix of routine and unusual events into a sample workspace. Follow each one through classification, decision and routing, and find the point where the cheap answer stops being enough.
session-3 xray, memory for any agentREMEMBER · 120 min 3/6 ›
What you’ll learn
- – Everything an agent observes or concludes is a memory cell: searchable by meaning, linked into a living graph, consolidated by reflection and knowledge loops, and let go when it stops mattering
- – Memory is laid out as in CoALA: issues and events as episodes, facts and patterns as knowledge, compiled rules as procedure
- – Over MCP, xray offers recall, remember and forget to any agent
- – Its code graph comes from Tree-sitter
Lab 3 · REMEMBER
Connect your own coding agent to xray over MCP. Remember a decision, recall it by meaning from a fresh session, then forget it and check that it is gone.
session-4 Tria, the teammateACT WITH PEOPLE · 120 min 4/6 ›
What you’ll learn
- – Tria knows the workspace, its services and how they call each other, its past issues and what the team learned from them
- – Tria speaks first when an issue opens or resolves, with the likely cause from what the workspace has seen before
- – You can mention @tria in a workspace chat so the whole room sees the answer, or talk to Tria by voice, and Tria asks before any change
- – How knowledge grows with every issue, and the messenger underneath
Lab 4 · TRIA
Open a sample issue, ask Tria for the likely cause in chat and by voice, and compare the answer with what the workspace remembers about similar issues.
session-5 The spot, an always-on agentACT ALONE · 120 min 5/6 ›
What you’ll learn
- – A spot watches your infrastructure through Triage and keeps working when you are away
- – It is a harness of memory, tools, skills and a sandbox, with connectors to Kubernetes, monitoring, cloud, messengers and MCP
- – The model can change; the harness keeps it the same agent
- – Each spot runs in its own sandbox on Agent Substrate, as an actor that suspends when idle, snapshots, and resumes with its tasks and conversation intact
- – It runs on Claude Code, a persistent session with Triage’s tools over MCP
Lab 5 · SPOT
Hand a spot a background task, such as preparing an RCA for each issue from last week, and hang up. Come back to the result and trace what the spot kept between visits.
session-6 ego and recursive self-improvementEVALUATE · 120 min 6/6 ›
What you’ll learn
- – ego samples live sessions from the OpenTelemetry traces your agents already emit, scores them, and alerts when quality drops
- – It needs no SDK and never re-runs the agent
- – The loop runs sample, score, detect, diagnose, correct and verify
- – When scores drop, ego explains why, hands the diagnosis to the agent over MCP, records what the agent changed and checks that the scores recovered
- – For an always-on agent, ego follows the work as one continuous thread
- – The wrap-up: putting the five parts side by side and choosing which one your team would start with
Lab 6 · EVALUATE
Read the findings for a sample agent, pick one, make a single change and watch the verdict. Compare the scores before and after.
The course concept and all course materials are the intellectual property of FWDAYS. Copying the content and materials of the course is permitted only with the consent of FWDAYS.
mentor
Denys Vasyliev
Principal Site Reliability Engineer / UK Global Talent Visa Holder
Event price
Attendee's ticket
Access to 6 online broadcasts (~15 hours)
Presentation and materials
Access to the course recording on the learning platform
Access to the chat in the TG, where the speaker will answer questions
Participant certificate (subject to homework completion)
Free month in Fwdays Club
Course participants receive a 10% discount on participation in Fwdays conferences
Available payment by installments from Monobank and purchase by installments from PrivatbankIf after the first lesson you realize that your expectations do not match, we will refund your money in full.
Group discounts are available for companies
Напишіть нам на academy@fwdays.com для прорахування вашої знижки.
Sign in
Sign in
Register with email
Register with email
Forgot password?