Hatz AI vs Formbricks

New models land here almost every week — Kimi K3 and GLM 5.2 Fast in the last few days, Claude Opus 5 before that, and Gemini 3.6 Flash, Flash-Lite, Inkling, and Grok 4.5 earlier in the month — all into one selector spanning chat, apps, agents, and workflows. Underneath that churn sits the MSP-specific layer where the durable work is happening: credit usage reporting across every tenant an MSP admin manages, Account Teams that scope which staff can open which customer accounts, and a searchable, exportable log of every AI invocation. Artifacts arrived in early July as a reusable output type — code, PDF, DOCX, PPTX, HTML — shareable by public link, and phone agents have picked up caller memory, business hours, blocklists, and department routing.

Two clocks run at different speeds. The model catalog turns over fast enough that individual additions carry little signal, and the guardrails shipped alongside Grok 4.5 — blocking publication of apps pointed at retired models and auto-healing existing ones to a successor — are a frank admission that churn has become an operational problem to absorb rather than a feature to advertise. The differentiated work is the control plane an MSP needs to resell this: usage visibility per tenant, access scoping per book of business, audit records the MSP can toggle per client. The Phone Agent is the one product surface being deepened rather than broadened, and it is now composable with the general agent runtime.

Expect the compliance log and cross-tenant usage reporting to grow into the oversight and chargeback story MSPs need to price this to their own clients, and expect phone agents to keep inheriting capabilities from the general agent runtime now that an existing agent can serve as a phone agent's brain.