Hey there 👋 I opened Komodor's launch yesterday looking for the new trick, the thing the agents could suddenly do that they could not do last month. It is not there. The whole pitch is about keeping agents grounded, giving them memory, gating what they touch, and holding the compute bill down. Then I hit the second paragraph and there was a Gartner line sitting in a vendor press release: more than 40% of agentic AI projects get decommissioned by 2027, on governance gaps, unclear payback, and runaway cost. Once I saw that sentence I saw it everywhere. WSO2 shipped a control plane. Workiva put agents inside its audit stack. Docusign is opening a governed way for any agent to touch a contract. None of it is a smarter model. All of it is insurance against a project getting killed next year. My bias, for what it is worth: this is the healthier phase, even if it reads less exciting. The demo era is what got 40% of these projects onto the chopping block. This week was the industry quietly agreeing that the hard part was never the demo.
The Big ThingKomodor shipped a platform for running the agents that run your opsStart with where the bottleneck went. Coding agents let teams push more changes in less time, and every one of those changes lands as an incident, a support ticket, a maintenance task, or a bigger bill. The load moved downstream, into operations. So the next agents you deploy are not the ones writing code, they are the ones keeping production alive, and on September 16 Komodor shipped a platform built for exactly that. The Komodor Agentic Operations Platform covers three jobs out of the box: AI SRE (troubleshooting, incident management, alert triage, reliability), Cost Optimization (cloud, observability, and Kubernetes spend), and AI Software Operations (change risk, CI/CD health, production readiness). Those ship as workflows running over 50 prebuilt specialist agents, skills, integrations, and MCPs. You can add or remove steps, reroute, and drop in your own agents. On top of that sits the part that matters more: a backbone to turn an existing script, runbook, or third-party agent into a governed one, running on the same context, memory, and knowledge graphs Komodor uses for its own. It is not the agents. It is the controls. Every custom agent runs under role-based policy that says who can invoke it and which credentials and tools it may use. Guardrails check the inputs, the tool calls, and the model's responses before the agent acts, and risky actions get held for a human to approve. Spending limits and a full audit trail keep the cost and the paper trail in one place. Komodor also shadow-tests new agent versions against validated scenarios before promoting them. Grade the numbers with your eyes open. The Gartner decommission projection is real, but you are reading it in Komodor's own release, positioned to sell the fix. The "60% of enterprise leaders already run agents in production" figure is a survey the vendor cites. The "50+ agents" and "up to 10x faster" are Komodor's own. Named customers (Nebius, Dell, Cisco, Prudential) are the more useful signal. Ship it? Deploy now if you run Kubernetes or production ops and want the backbone from day one. If you do not, copy the four controls anyway: guardrail before the agent acts, a human gate on risky moves, a per-agent spend cap, and an audit trail. That is the shape every operational agent needs, whoever builds it. Source: Komodor press release via GlobeNewswire, "Komodor Launches Agentic Operations Platform," Sep 16 2026. Gartner and survey figures are cited within that release; graded as vendor-supplied.
Tour de Headlines🏛️ WSO2 shipped the open answer to agent sprawl. Its Agent Manager reached general availability on September 15 as an open control plane, released under the Apache 2.0 license, that governs agents across any framework, model, or deployment. It hands each agent an identity, enforces policy at scale, and adds full-stack observability, whether the agent runs inside your walls or somewhere else. The whole pitch is sovereignty: deploy it anywhere, keep the agent data on your side, and do not get locked into one vendor's stack just to get governance. This is the same governance problem IBM and Salesforce are selling control planes for, with one difference that matters to a lot of teams. You can read the source and run it yourself. It went to beta back in May, so the September news is the GA, not the idea. 📋 Workiva put an agent builder inside the audit file. At its Amplify conference the company launched Agent Studio, a no-code way to build, customize, and schedule agents inside its governed reporting platform, grounded in a company's own filings and data. Alongside it came three agentic solutions aimed at work that has stayed stubbornly manual: BEA surveys, US Census surveys, and Country-by-Country Reporting. In a regulated shop the guardrail is the product. An agent prepares the disclosure, a human signs off before it acts on anything regulated, and the output stays traceable enough to defend to an auditor. The pattern travels to any high-stakes vertical: let the agent prepare, keep a person on the approval, and attach the trail before, not after. 🏢 Hancom framed agents as a workforce you manage. Nomadian, from the 36-year-old Korean office-software maker that rebranded around an "agentic OS," lets you hand over a task in one sentence, splits it across role-based agents laid out in a 3D virtual office, and keeps final approval in human hands. Fair warning, this one is a watch, not a deploy: the US beta lands in December, and paid subscriptions are a 2027 thing. It earns a mention anyway because of the framing. Job titles, a visible office, explicit sign-off steps. That is how multi-agent orchestration gets sold to a buyer who does not care what a model is, and the packaging problem it solves is one half the tools above still have not.
Sponsor Your team's calls are full of signal. Measure it. RapportScore reads the calls your team already records and scores how your people communicate, then turns it into coaching they can use. Stop guessing whether a rep is building trust or talking past the buyer, and see it in the numbers. See your team’s score → |
Tool of the Day🔧 Docusign MCP ServerGive any MCP-speaking agent a governed way to draft, send, and track agreements. If your agents touch contracts, this is the one to wire up this week. Docusign is taking its MCP Server to general availability on September 30, which makes agreement actions callable natively from Claude, ChatGPT, Gemini, Copilot, Slack, or any MCP client. It ships with account-level admin controls, multi-region infrastructure, and multilingual support, and its Iris engine feeds the agent the context of past negotiations, accepted terms, clauses, and company policy so it is not drafting blind. The server has carried a beta label since February, so you can build against it today and be ready the day it flips. The one-line reason to care: this hands an agent the power to move an agreement without handing it the keys to your whole account. Read the Docusign announcement
Worth a Click- Komodor Agentic Operations Platform. The September 16 launch in Komodor's own words: 50+ prebuilt ops agents, a backbone to bring your own, and the governance stack of guardrails before an agent acts, human-gated risky actions, spend caps, and an audit trail. (globenewswire.com)
- WSO2 Agent Manager (open source). The Apache 2.0 control plane itself. Read the source, self-host it, and govern agents across any framework or model without locking into one vendor. (github.com)
- Docusign Agreement Layer for the Agentic Enterprise. What goes GA on September 30: the MCP server that makes governed agreement actions callable from any agent, with admin controls and Iris context. (prnewswire.com)
Call it the anti-decommission stack. That Gartner line, more than 40% of agentic projects killed by 2027, is the sentence every launch this week was written against. Komodor sells the production backbone. WSO2 sells the open control plane. Workiva sells the audit trail. Hancom sells the human-approval framing. Docusign sells the one governed action an agent is allowed to take. Strip the logos and none of it is a cleverer agent. The thing on the shelf now is a reason your agent project is still funded after the next budget review. That is a less thrilling product than a smarter model, and it is probably the one that keeps more of these projects alive. See you tomorrow.
— Ron |
