AI headlines for business this week (Sep 28, 2026)
Support publishes customer-facing handoff SLAs, security audits break-glass on agent consoles, finance turns on workflow chargeback, legal templates ediscovery for agent chats, and SRE teams drill model provider outages.
Third week of January and customers are asking the same question in every vertical: when does the bot stop and when does a person show up? Teams that cannot answer in writing are losing trust faster than they are saving handle time.
Last week's IAM, ERP tags, and escalation headlines wired identity and cost. Pair that week with the IAM permissions deep dive. This week is about what customers see, what legal can find, and what happens when the model API blinks.
1. Support publishes customer-facing agent handoff SLAs
CX teams post plain-language SLAs on help centers: max wait in an agent loop, how to request a human, and response targets after handoff. Status pages link to the same numbers support uses internally.
Why it matters for business: vague "AI assisted" footers do not cut it. Read the deep dive on support escalation SLAs and approval queues.
2. Security audits break-glass access to agent admin consoles
Identity teams review who can change prompts, tools, or keys without a ticket: shared admin logins get retired, break-glass accounts get time limits, and every override logs to SIEM.
Why it matters for business: console access is production access. Pair with IAM-mapped tool permissions and prompt promotion pipelines.
3. Finance activates chargeback on workflow-tagged agent spend
Controllers flip on department chargeback for agent API lines that already carry workflow codes in the ERP. Budget owners see variance in the same report as SaaS seats.
Why it matters for business: tagged spend only helps when someone owns the number. See spend dashboards and kill criteria.
4. Legal adds agent chat logs to ediscovery templates
Counsel updates standard discovery request language to include agent transcripts, tool call logs, and retention windows. IT gets a checklist of which systems export in native format.
Why it matters for business: surprises in litigation are expensive. Align with transcripts and legal hold and year-end transcript exports.
5. SRE teams rehearse model provider outage runbooks
Reliability engineers run tabletop drills for provider latency spikes: degrade to cached answers, route to humans, or pause agents that cannot run safely without the model.
Why it matters for business: your SLA is only as good as your fallback. Compare eval gates before promotion and runtime sandboxes.
What I would do this week
- Publish one customer-facing handoff SLA for your busiest agent channel.
- Remove one shared admin login from your agent console.
- Send workflow-tagged agent spend to one department owner as a trial chargeback.
- Read the escalation SLA deep dive.
Want blog updates? Join the notify list.
Matt Potter · Swift Media