June 2026 carries on the spring momentum, but the tone has shifted. Until now, agentic AI was being launched: SW1 at SolarWinds in April, the first SRE agents at New Relic in February. In June, it moves into the everyday products: an MCP server inside Centreon, an AI chatbot inside PRTG, an autonomous SRE at New Relic. And, for the first time, it is also the month when the first study that coldly quantifies the collateral damage arrives. At Sensor Factory, we read it across the board: releases that are maturing, AI settling into the tool itself, and a budding clarity about what it costs.
Releases & LTS: the machine keeps turning
On the Zabbix side, two maintenance releases on 2 June: 7.0.27 LTS and 7.4.11, on both active branches. On 29 June, a 6.0.47rc1 release candidate appeared on the legacy 6.0 LTS branch, proof that this branch is still maintained for estates that have not migrated yet. Nothing on 7.6 or 8.0 this month: 7.0 remains the recommended target for new deployments.
At VictoriaMetrics, v1.146.0 shipped on 22 June: new stream aggregation options (deduplication and a new output type), prioritisation of recent data on the query path, improved cluster sharding for vmagent, and several fixes including a metadata corruption in cluster mode. But the operational item to remember concerns the Kubernetes operator: v0.71.0 (12 June) should be avoided, a bug stops metric emission beyond a certain number of tracked objects. v0.72.0 (15 June) fixes it, adds validation webhooks for the Prometheus Operator CRDs and resolves the status-collector deadlock beyond 250 objects. If you run the VM operator in production, jump straight to 0.72.0.
Centreon is preparing its next LTS. The preview of Centreon Infra Monitoring 26.10 LTS announces support for Enterprise Linux 9 and 10 (RHEL, AlmaLinux, Oracle Linux) and Debian 13, dropping EL8 and Debian 12. Estates still on those OSes have a modernisation project to plan before the switch. We come back to this below, because the preview also hides an AI novelty.
Agentic AI moves into the products
This is the defining thread of the month. Three vendors are shipping AI features built into the tool, not bolted on the side.
New Relic held its New Relic NOW event on 23 June and unveiled Autopilot, an autonomous SRE that investigates and resolves incidents from production data, with Kubernetes and Kafka specialists. Alongside it, Ground Truth opens AI-optimised access to New Relic data, governed by authentication and RBAC. General availability is announced for late July. We move from « AI that assists » to « AI that acts », with, at least on paper, a governance layer.
Paessler shipped PRTG 26.2.120 on 18 June in the stable channel: three new SNMP sensors (Custom String v2 with numeric extraction via regex, Printer v2, RMON v2 rewritten for the Multi-Platform Probe), a new « Installation Report » (licensing, system health, recommendations) and, above all, the arrival of MAX, the in-house AI chatbot, on the welcome page of the new UI for in-app help and troubleshooting. PRTG stays true to its promise, simple agentless monitoring, but is starting to dress the experience with a conversational layer.
Centreon, precisely, slips into the CIM 26.10 LTS preview an open-source CIM MCP Server (published on GitHub): it lets you query monitoring data in natural language through an AI assistant. This is exactly the topic we covered in our article on MCP tunnels: the MCP protocol is becoming the standard entry point to wire an LLM onto a monitoring tool. When a long-standing vendor like Centreon ships it natively, the signal is clear.
Even IP-Label / Ekara joins the movement, on the editorial side: its 24 June article on generative and agentic AI in monitoring shows the topic now shapes the whole ecosystem’s discourse, including digital-experience specialists.
…and the clarity comes with it
June’s most interesting counterpoint comes from the same New Relic. On 10 June, the vendor published its « State of AI Coding » report, and the figures are worth pausing on: 94 % of leaders judge AI-generated code higher quality at review time, yet 78 % see a rise in incidents in production after deployment, and 82 % suffered at least one production outage tied to AI code over six months. The report introduces the notion of « agent debt », the equivalent of technical debt, but generated by agents that quickly produce plausible, poorly understood code.
The collision is striking: the same vendor launches an autonomous SRE and documents that code-writing AI breaks production more often. That is not a contradiction, it is maturity. After the 2025 enthusiasm, 2026 is starting to measure. And the practical takeaway is clear: the more we delegate to agents, whether they write code or remediate incidents, the more critical the observability of what they do becomes. Agentic AI does not reduce the need for monitoring, it shifts it.
Moves & weak signals
SolarWinds made two corporate announcements: a partner program overhaul on 4 June (tier-based discounts, a Fast Track program, effective 3 August), and on 17 June the appointment of Justin Henkel as CISO, a solid security profile (ex-OneTrust, US Air Force intelligence), a choice consistent with the vendor’s AI push and the governance need that comes with it. For the record, their SW1 agent dates from April, not June.
On Centreon security, no new bulletin in June, but the 28 May one (two CVEs, CVSS 6.6, shell injections in legacy PHP code, fixed in Centreon Web 25.10.12 and 24.10.25) is still worth applying if it is not done. Also worth noting in the 24 June connector delivery: a NRPE breaking change (check_centreon_nrpe3 replaced by check_nrpe) and native support for SNMPv3 macros in host templates, two items to fold into your update procedures.
On the niche-partner side, the month is quiet on the product front. Mr Suricate mostly communicates about its presence at VivaTech 2026 (10 June), no June-dated Magui.ai news. M81 (IBM i monitoring) and Martello published nothing new. As a reminder, Vantage DX has been End of Sale since late 2025, a hard fact to know if you operate Microsoft Teams or UC monitoring. Better to say it than to invent an announcement: not every month is equal, and honest tech-watch also means flagging the silences.
The cross-cutting read
Three observations we draw from this month of June:
- AI leaves the press release and enters the binary. MCP Server at Centreon, the MAX chatbot in PRTG, Autopilot at New Relic: we are no longer talking roadmap, we are installing versions. The CIO question is no longer « do we go for it? » but « what scope do we hand over, and how do we observe it? ».
- Governance becomes a product argument. RBAC-governed Ground Truth, a CISO appointment at SolarWinds, New Relic’s « agent debt » report: the ecosystem is internalising that delegating to agents without guardrails means creating invisible debt. The « AI everywhere » talk of 2025 nuances into « governed AI » in 2026.
- Operational discipline still rules. A Zabbix LTS maintained down to 6.0, a VictoriaMetrics operator version explicitly discouraged, a Centreon preview that forces OS modernisation: behind the AI, the fundamentals, namely versions, supported OSes and migration windows, still decide real-world reliability.
Observing the agents: our two in-house bricks, SenHub agent and Toise
If the 2026 stake is observing what agents do, we may as well own it: this is exactly the conviction behind our two open-source, OpenTelemetry-native projects, SenHub agent and Toise.
SenHub agent is our collection agent, designed OTel-first: it speaks OpenTelemetry end to end, stays light to deploy, and covers infrastructure and database monitoring. The idea fits in one sentence: standardised, portable collection, with no proprietary lock-in on the data.
Toise is our entity-and-relationship graph. Where a classic tool collects metrics, Toise maps the entities of an information system, their history and their dependencies, and answers impact questions: « if this component goes down, what does it touch? ». It aligns with the OpenTelemetry « Resources and Entities » SIG work (entity-events), the emerging standard for describing entities and relationships in observability. When an AI agent remediates an incident on its own, knowing which entities it changes and what that impacts is no longer a nice-to-have, it is the condition for trusting it. That is precisely what Toise makes observable.
What we do with it at Sensor Factory
Concretely, these announcements guide our recommendations. We advise affected clients against VictoriaMetrics operator v0.71.0 and plan the move to 0.72.0. We document the NRPE breaking change and the SNMPv3 macros in our Centreon procedures, and we check that the May security bulletin has been applied. We are starting to scope OS modernisation projects ahead of CIM 26.10 LTS. And above all, we evaluate the agentic bricks, Centreon MCP Server, New Relic Autopilot, for what they actually enable, keeping the « agent debt » report in mind: an autonomous agent is only worth it if you can observe and bound what it does.
If you want to take stock of your stack in light of these changes, LTS migration, AI-agent governance, OS modernisation, let’s talk.
This tech-watch is a monthly format. The next edition will cover July 2026. To receive it automatically, the newsletter sign-up is at the bottom of the page.
