Three vendors ship unfinished products while support ends

The deadlines are firm, the products are not.

Three announcements from yesterday, read one after the other. Each one hands you a date somebody else chose, and each date points at something that is not finished being built. Two of them arrive with a retirement notice attached.

Advertisement

1. Varonis shipped three products three weeks after agreeing to be sold

On Monday Varonis added three capabilities to its Atlas platform: Browser Protection, Endpoint Protection and an AI Gateway. All three are available now.

On 16 October Proofpoint agreed to buy the company for about $6.3 billion, $49.75 a share in cash, with closing expected in the first half of 2027. A write-up the next day put the end-of-life date for the self-hosted product at 31 December 2026. On 20 October Varonis connected itself to Collibra to tag sensitive data inside the catalogue, and Forrester called the deal an integration test.

A company that has just been sold still shipping product is normal, and I read nothing into it. The date I read twice is 31 December. The version you host inside your own walls loses support while the firm that promised to support it is mid-sale, and whoever signs a renewal in January is really negotiating against Proofpoint's roadmap. That is an ownership change arriving before the paperwork does. I would get the migration commitment on paper this month, while the people who made it still work for the company that made it.

Read the source →

2. Microsoft's ontology is in preview and the old one has a retirement date

A consultancy walked through where Fabric IQ actually stands, and the dates are the readable part. It became generally available inside Copilot Chat and Copilot Cowork on 28 September, at no extra token cost. The new ontology experience, the layer that defines what a customer or a margin is before an agent answers about one, is still in preview as of yesterday.

The previous ontology experience retires on 31 January 2027.

Microsoft's own tests have agents grounded in an ontology giving 2.2 times more answers it rates excellent, with 30% fewer tool calls. Gartner's caution in the same piece: ontologies generated out of existing semantic models come out incomplete or siloed, and subject matter experts are still the ones who have to validate them. Microsoft is also co-developing a standard called Apache Ossie with Snowflake for passing this kind of metadata between platforms.

The trail around it is a crowd. Gartner wrote on 19 October that these updates make ontologies operational, with AI-assisted ontology creation in public preview. Futurum, five days earlier, read the same announcements as the semantic model becoming a control plane. On 5 November Zoho made the same bet from the other end of the market, context before agents.

The thing meant to tell an agent what a customer is, is itself in preview. What I would not do is move a company's definitions onto it on the strength of a vendor's own grading of its own answers. An ontology is not a feature you switch on, it is a maintenance commitment, and Gartner just named who pays for it: your subject matter experts, every quarter, for as long as the business keeps changing. My expectation is that most of these go stale within a year, because the definitions get loaded by a platform team and then owned by nobody. So I would put a named person against every governed metric before any of it loads, and a metric nobody will put their name to does not go in.

Read the source →

Advertisement

3. Salesforce says the price of an agent is still a work in progress

A read of the numbers published yesterday has Agentforce past $1.5 billion in annual recurring revenue, up 240% on the year, and Agentforce plus Data 360 at about $3.9 billion, up 210%. Half of Agentforce bookings are existing customers buying additional credits. The chief executive says the pricing model remains a work in progress and that the company is “trapped in some ways in old per-user pricing models”. Non-current debt is at $39.3 billion, against $10.4 billion before.

Half the bookings being top-up credits is the figure I would carry into my own finance meeting. Read as expansion it is a growth story, and read as a budget it says the first estimate was wrong, by enough that customers are back inside the same year buying more. When the vendor's own chief executive says the shape of the price is unsettled, cost control is not something you can delegate to the contract you signed, because the credits get bought mid-quarter and the budget was set before anyone knew what an agent run costs. I would set a hard quarterly ceiling and an alert at 70% of it, and treat the first breach as a scoping error rather than a success.

Read the source →

Advertisement

Also today

• Morgan Stanley agreed to acquire Vinyl Equity, a digital-first, SEC-registered transfer agent. (7 min read)
• India's consent manager framework under the data protection rules carries a November 2026 deadline for data fiduciaries. (12 min read)
• ChatGPT's growth has rebounded while Claude and Gemini trail on consumer share. (2 min read)

First MoveThe radar behind the autonomy company

The number repeated everywhere this week was $28.8 billion, Shield AI's valuation after a $2.5 billion raise. The thing it is paying up to $5.3 billion for is a radar company. Not a model, not a dataset, not capacity: the antenna that decides whether the autonomy is looking at a bird or a drone. Echodyne's own business is almost unwritten about, while every write-up of it is about the people buying it. That gap is the whole point of looking one hop behind a deal.

What is the sensing layer under your agents, and who owns it?

My call: by 30 June 2027, at least one US-listed enterprise software company announces the acquisition of a hardware sensing or telemetry maker.

Read the deep dive →