SAP puts its agent containment work into OpenShell
SAP's four contributions to the OpenShell agent runtime are containment and operability engineering, and the runtime is free only through October.
Topic coverage
AI & Agents
Source-linked reporting and practical patterns organized around a shared enterprise applications question.
Latest reading
Salesforce's three tiers meet a five percent upgrade rate
Core, Advanced and Max list at $195, $395 and $550 with named credit pools, and five percent of Sales and Service users have moved up so far.
Salesforce's named agents are only as safe as your permissions
Rebecca Wettemann's read on Salesforce's named agent roster turns on one condition an admin can test, that permissions are correct today.
Salesforce names six trust layers and two need proof
Salesforce's newsroom post lists six trusted capabilities behind its agent platform. Trusted agency and trusted actions are the two an architect has to test.
SAP says autonomous already sits in the 2023 release
SAP's recap of its own podcast quotes one executive saying autonomous capability is available now on the 2025 and 2023 releases. We cannot verify that.
Workday screening plaintiffs move to certify a class
The named plaintiffs moved on September 14, 2026 to certify a class against Workday itself, and the briefing runs to a March 9, 2027 hearing.
Work IQ lets Copilot write back to your source records
Work IQ public preview opens September 30 with governed actions that write changes back to source records inside the permissions your users already hold.
Salesforce names a headless meter and leaves the price blank
Salesforce's August 31 rate card carries a Headless Platform Interaction usage type priced TBA, and the two narrower rows it replaced are gone.
Dynamics 365 ERP MCP server ships ahead of its controls
Microsoft's Dynamics 365 ERP MCP server is generally available, while the role-based access controls written for agent use are still described as coming.
Oracle Health says it intends to reach Fusion finance
A vendor press release states an intention to connect reimbursement workflows to Fusion financials. Nothing named in it is generally available yet.
ServiceNow's Brazil release notes arrive with a date
ServiceNow's Brazil release notes are live and dated, with a documented September 24 Early Availability date and roughly 90 product areas changed.
Certinia's agent count and the move to fixed-bid work
Certinia went from 10 packaged agents in April to 24, and the bigger change is Salesforce-native firms quoting fixed bids instead of hours.
Dynamics 365 Field Service agent will apply its own schedules
Microsoft's Field Service scheduling agent is moving to recurring runs that write optimised bookings onto the board with no dispatcher starting them.
SAP's ITOCHU example picks a hard general ledger job
An SAP column names ITOCHU applying AI to general ledger posting for raw-material trades. No result, scale or date has been published yet.
An analyst says Now Assist is pulling tier upgrades
Cantor Fitzgerald raised its ServiceNow target on client checks describing tier moves for Now Assist. The renewal problem that creates is yours.
Oracle's HCM Leader claim is really an agentic pitch
Oracle's own release says Fusion HCM moves beyond surfacing insights to take on the work. That framing, not the badge, is what you meet at renewal.
EssilorLuxottica's Joule glasses leave a consent question open
SAP says two Joule smart glasses pilots at EssilorLuxottica were built in two weeks. The shop-floor one recognises customers, and consent goes unmentioned.
Buyers expect headless Salesforce access to cost more, not less
Buyer research on 20 interviews found most already reach Salesforce outside its UI, and three quarters of those modelling it expect to pay more.
ServiceNow adds two flow debugging tools, one you can test
Input Modification ships with Brazil and reaches personal developer instances on September 24. The Troubleshooting Agent needs Otto and Now Assist for Platform.
Salesforce says every job now has a builder component
Salesforce Ben reports almost 20,000 staff are in roles that did not exist a year ago, and that every role now screens for a builder component.
SAP's accruals agent stops at the approval step
SAP's new accruals agent drafts journal entries and shows the policy rule and the math behind each one, but an accountant still decides what posts.
ServiceNow's kill switch reaches into Okta, GCP and AWS
ServiceNow's new agent kill switch orchestrates credential revocation across Okta, GCP and AWS. What the post does not say matters more.
Workday 2026R2 is in your tenant, Agent Passport included
Workday 2026R2 reached production on a Saturday. The automatic and optional change split is Monday's problem, and Agent Passport deserves harder questions.
The TSA number inside Salesforce's Missionforce year one
Salesforce says TSA's Ace agent resolves 96% of roughly 100,000 monthly traveler conversations without escalation. That rate deserves a second look.
Testing Oracle's case against hollowed-out apps
Oracle argues AI accelerates packaged applications rather than replacing them. We test which parts of the app layer actually hold.
Power Platform agents can now author security roles
Microsoft's app-builder skill reached GA covering security roles and business rules. What that changes for Power Platform review, ALM and connectors.
Salesforce's Koa puts CRM reasoning inside the model
Salesforce claims three times fewer errors on CRM actions for Koa. That figure comes from Salesforce's own CRM benchmark, not an independent one.
Salesforce's Agentic Work Unit is a budgeting problem
Salesforce bills agent work in Agentic Work Units, and one business outcome can consume wildly different numbers of them. Your renewal needs a forecast.
Oracle reported agent executions, so we did the math
Oracle says 2,300 agents ran 3.5 million times last quarter. Divide one by the other and the number you get is worth a look before you plan your own rollout.
SAP turned AI token spend into a budget line
SAP now plans AI token spend inside its normal finance cycle after employee and agent usage became a triple-digit-million-dollar risk.
ServiceNow's shift zero and the AI tools CIOs can't see
Three quarters of CIOs told diginomica they can't see every AI tool in their systems. ServiceNow's shift zero pitch depends on data they don't have.
Siemens' Agentforce rollout is a governance test, not a win
Siemens is routing 2,500 monthly leads across 132 countries through two Agentforce agents, and the real test is quality control at that scale.
ServiceNow's newest agents are Experian's, not its own
ServiceNow will run Experian's risk and identity agents on live loan and claims decisions, but Experian built and controls them.
Claudeforce's 37 skills turn a preview into a permissions problem
Salesforce put a number on Claudeforce at Dreamforce 2026, 37 sales skills running through an MCP server in Claude, and that number is a permissions question first.
AIforce claims to replace the UI. The admin's work just relocates.
Salesforce framed AIforce at Dreamforce 2026 as AI replacing the CRM interface. For the admin who built what's underneath it, the question is where that logic goes next.
ServiceNow's autonomous security specialists go live
The autonomous security specialists ServiceNow previewed in June reach general availability today, and the checklist for trusting them is still thin.
ServiceNow's saved-a-month claim needs a formula
Raleigh's CIO says ServiceNow AI saved a month of time. The number only means something once you know whose month, measured how.
Job-ready Agentforce agents need a defined handoff between jobs
Salesforce's job-ready Agentforce agents each own one function. When orchestration passes work between them, your team must define what crosses and who owns it.
ServiceNow MCP servers still run on your roles and ACLs
ServiceNow's MCP servers keep ACL and role enforcement in place, so what an agent can do is decided by the role model your instance already has.
Dynamics 365 Activate needs a risk register before it runs
Microsoft's Activate preview automates Salesforce-to-Dynamics 365 discovery. The ownership, mastery, integration, and rollback questions come first.
What SAP support numbers tell an architecture team after go-live
SAP's September 10 services update counts Joule users and success plans. After go-live, the number an architecture team needs is in its own support queue.
Salesforce's Trusted Context is the real boundary for agents
Salesforce's six-part enterprise AI model lists Trusted Context first. Our read: what an agent can see and understand is the limit of what it may do.
What an Agentforce agent needs before it can run for weeks
Salesforce's new long-horizon runtime lets Agentforce agents pursue goals for weeks. These are the run-level controls the post leaves for you to ask about.
ServiceNow AI Gateway puts agent policy at the tool call
ServiceNow's AI Gateway enforces policy as an agent invokes an MCP tool. Here is which rules belong at that layer and which still belong to the record.
SAP's Embodied AI Jam turns master data into a safety question
SAP put drones and humanoids on Joule at its first Embodied AI Jam. Our read: every use case rests on an equipment or routing record a person keeps accurate.
Dynamics 365 Activate and what each customization was for
Microsoft's Activate preview surfaces Salesforce customizations before the first workshop. The first review has to ask which need each one met.
What Salesforce's AI Control Plane needs to explain one action
Salesforce's AI Control Plane will observe agent behaviour and outcomes. Explaining a single action needs a per-call record the post never describes.
PPCC 2026 agent demos need a process owner in the room
Microsoft's PPCC 2026 preview asks attendees to bring one process. The stage agent will work either way. What decides the pilot is who owns that process.
New Salesforce editions move the buying line into architecture
Core, Advanced, and Max bundle Slack, Tableau Next, and Flex Credits into every Agentforce seat, so the tier choice now belongs in architecture review.
The five-part test we'd run on SAP's Embodied AI Jam pilots
SAP's Embodied AI Jam sent teams home with use cases and architectural proposals. Here is the five-part test that separates a robot demo from a process.
New Salesforce editions put agents past the intake gate
Core, Advanced, and Max bundle Agentforce into every seat, so the purchase request that used to trigger a governance review never gets filed.
PPCC 2026's procurement example gets the approval step right
Microsoft's PPCC 2026 procurement example sends only higher-risk exceptions to a person. Most approval steps in a live tenant would fail that test.
Winter ’27 longer-running agents need a named lifecycle owner
Salesforce says Winter ’27 agents run whole workflows across channels. A run that lasts days outlives your change window, so someone must own it.
What a warehouse robot needs from SAP after the Embodied AI Jam
SAP's Embodied AI Jam listed warehouse automation without naming a system. Here is the task data, write-back, and exception path a floor team needs first.
The review questions a faster AI Agent Studio still needs
ServiceNow's reworked AI Agent Studio goes from Advisor opportunity to running agent on one screen. These are the review questions to ask before the click.
PPCC 2026's procurement agent needs one record it writes to
Microsoft's PPCC 2026 preview runs a procurement request through Copilot, an agent, a workflow, and a reviewer. We'd ask which Dataverse row holds the trace.
What the new Salesforce edition prices leave out of the bill
Salesforce's Core, Advanced, and Max editions carry clear seat prices. The Flex Credit pool, bundled overlap, and migration labour decide the real bill.
Winter '27 puts governed CRM data under every agent
Salesforce's Winter '27 post rests autonomous workflows on governed CRM data. Here is what governed has to mean before an agent runs a case end to end.
Winter ’27 cross-channel agents redefine a closed case
Salesforce says Winter ’27 agents carry a case across voice, Slack, and email. Closed now means every channel is settled, and your status field can't say that.
Workday's 5,500 agent customers, and what the count leaves out
Workday's Q2 release says more than 5,500 customers use at least one organic agent. The count says the switch is on. It says nothing about the work.
Winter ’27 gives you six weeks, so build the calendar today
Salesforce’s Winter ’27 post is dated August 31 and goes live October 12. Here is how to turn its fifteen items into owners, sandbox dates, and a freeze.
Claudeforce prebuilt skills carry someone else's sales process
Salesforce in Claude ships 37 prebuilt sales skills. Each one encodes a definition of deal health and pipeline review that your org may not share.
Workday's agent quarter still runs on business process rules
Workday says AI drove a quarter of new ACV. The condition rules and approval chains already in your tenant are what make those agents safe to switch on.
Claudeforce should leave permission ownership inside Salesforce
Salesforce says an admin connects Salesforce in Claude once and permissions are managed centrally. Whose identity each action runs under is the question to ask.
Workday's Q2 momentum belongs in your architecture plan
Workday's fiscal 2027 Q2 release reports a $27.4 billion subscription backlog and AI in a quarter of new ACV. Here is how to write that into a plan.
Claudeforce solves the read path and leaves you the write path
Salesforce in Claude reads CRM and Slack data through one admin connection. Governing the writes, where business rules and side effects live, is the hard part.
The Workday platform changes hiding in the Q2 results
Workday's fiscal 2027 second-quarter release lists a run of product changes under the revenue line. Here is how a tenant owner should sort them.
What Claudeforce has to preserve when sellers work in Claude
Salesforce says Claudeforce routes actions through its rules. Here is what the trust boundary must still carry when a seller never opens Salesforce.
What the Claudeforce open beta asks of a Salesforce team
Salesforce says Claudeforce moves from select pilots to open beta in September 2026. What a Salesforce team should settle before sellers connect Claude to CRM.
Analysis and architecture
What an ERP vendor's agent claims are actually worth
Every finance suite now sells agents that reason and act. Four questions about scope, identity, evidence and doubt tell you which ones are deployable.
What a CRM is for once capture stops being manual
Automatic capture ends the old CRM bargain. Recorded volume rises, the average record matters less, and the reason behind a decision becomes the scarce thing.
How far to extend Oracle Fusion before you build outside
Every extension inside Fusion is a standing commitment to retest four times a year. Exhaust configuration first, and keep logic with its own release cycle outside.
Grounding Joule starts with the authorisation model
An assistant grounded in SAP data is only safe if it inherits the authorisation check rather than retrieving broadly and filtering the results afterwards.
Deflection design for ServiceNow customer service portals
An employee who fails at self-service tries again tomorrow. A customer may just leave, and may hold a contract saying you owe them an answer.
Salesforce knowledge an agent can safely answer from
A human reads around a bad knowledge article. An agent answers from it, which turns knowledge quality into an accuracy problem with a blast radius.
The platform admin job when agents do the building
Agents can build the field, the flow and the report. What they cannot do is know why your configuration looks the way it does, or which request to refuse.
Why agents in Workday need a stricter governance bar
An agent that changes a worker record carries risks a service desk agent never does, and HR deployments need a governance bar set well above the usual checklist.
What still earns a premium in platform consulting
Agents are absorbing the configuration work, and the premium is moving to judgement, boundaries and saying no.
What case deflection actually saves, and what it costs
Deflection metrics count three different outcomes as one success, and automation takes the cheapest cases first. Here is how to measure what you really saved.
Agentforce or Copilot Studio when you run both stacks
For an organisation running both stacks, the agent belongs where the data and the permission model that has to govern it already live.
What actually changes in a rep's week when agents do the legwork
Research, activity logging and quote assembly really do shrink. The judgment calls do not, and checking the agent's output before a customer sees it is new work.
What I'd test before trusting Ask Oracle with permissions
Oracle says Ask Oracle respects your NetSuite permissions. Two months into rollout, here's the edge case worth testing yourself.
Designing the human review step for Copilot Studio agents
Name the decision a person keeps, show the reviewer the source and the proposed change, give the queue an owner, and write the answer back to the record.
AI agents need a pause button someone can actually reach
A pause control with an owner, a written reason, and a resume record should be an ordinary operating control for Copilot Studio and ServiceNow agents.
Related guides
Turning on Oracle Fusion agents without losing the audit trail
Scope, identity and a realistic approval step decide whether a Fusion agent rollout stays reviewable. Capture the baseline and the off switch first.
What Agentforce actually is, and where it stops
Agentforce agents are handed a job rather than a script, bounded by topics and instructions, and they see exactly what the user they run as can see.
The readiness check before an agent pilot grows
Before a Copilot Studio or Agentforce pilot gets more users, check it has four named owners, a visible failure path, and a review date.
What your ServiceNow CMDB needs before an agent reads it
An agent assessing change impact trusts every CI and relationship it finds. Fix class ownership and identification rules before it goes live.