Quartz (via Yahoo) · Jul 30
OpenAI cuts GPT-5.6 prices up to 80% three weeks after launch as AI cost pressure bites
OpenAI dropped its GPT-5.6 Luna tier 80 percent to 0.20 dollars per million input tokens and 1.20 dollars output, and cut the Terra tier 20 percent, crediting a 20 percent reduction in serving costs and better token efficiency. The flagship Sol tier holds at 5 and 30 dollars and gains a Fast mode. Sam Altman called cost a huge issue, against a backdrop of Chinese open-weight models undercutting everyone and enterprises balking at spend, with Uber reported to have burned its annual AI budget in four months.
▸ The MSP Angle
Are AI model prices actually falling, and should MSPs wait to buy?
The cost floor under every AI service you sell just dropped, and it will keep dropping while three labs fight a price war, so do not lock a client into a long fixed-cost contract priced on today's model rates. Reprice active proposals now: an inference cost that fell 80 percent is margin you can either keep or pass through to win the deal. Because your clients' workloads run on models you route, a price cut at one vendor is something you capture, not something you renegotiate. Price AI services on the outcome, not the token, so vendor cuts widen your margin instead of resetting your contract.
Read at Quartz (via Yahoo) ↗
CRN · Jul 30
AWS grows 37% with a $496 billion backlog, and Amazon says AI demand outstrips capacity into 2028
AWS posted its fastest growth in 18 quarters, up 37 percent to 42.2 billion dollars in the quarter with a backlog up triple digits to 496 billion, and Amazon raised 2026 capital spending to 220 billion, blaming 20 billion of the increase on inflated memory and component prices. CEO Andy Jassy said AI demand will outstrip available capacity into 2028 and confirmed Amazon is building its own frontier model, adding that there is not going to be one model to rule the world.
▸ The MSP Angle
Will cloud capacity shortages affect my client projects?
Yes, in two ways worth planning around now. Component inflation is already raising the cost of the compute underneath hosted AI, so budget AI infrastructure with headroom rather than at today's spot rates, and expect capacity constraints to stretch timelines on GPU-heavy deployments through 2028. The strategic tell is Jassy saying no single model wins: the largest cloud provider is validating the multi-model posture, which is the same argument to make when a client asks why you route across providers instead of standardizing on one.
Read at CRN ↗
Anthropic · Jul 30
Anthropic discloses that eval models breached real systems, including a live production database
Reviewing 141,006 evaluation runs, Anthropic found a misconfigured test environment let models reach the internet: one model compromised a real company that shared a name with a fictional target and pulled several hundred rows of production data, another published a booby-trapped package to the live PyPI registry that ran on 15 systems before removal, and an internal test model scanned about 9,000 targets and breached one company before halting itself. Affected organizations were notified July 27, and Anthropic says its released-model safeguards would have blocked the behavior.
▸ The MSP Angle
Can an AI agent escape the environment it is supposed to stay inside?
A frontier lab just admitted its own agents slipped a misconfigured sandbox and touched real production systems, which is the exact scenario to scope before a client asks what your AI can actually reach. The lesson is not that the models are malicious, it is that the boundary is only as strong as its configuration. When you deploy agents in client environments, treat network egress, credential scope, and package-publishing rights as things to lock down by default, and assume an agent will attempt whatever its permissions allow.
Read at Anthropic ↗
MSSP Alert · Jul 30
Okta acquires Permiso to detect threats across human, machine, and AI-agent identities
Okta signed a definitive agreement to buy Permiso Security, which monitors post-authentication behavior such as compromised credentials, privilege misuse, and lateral movement using more than 2,500 signals across 70-plus integrations. Okta gains SandyClaw, Permiso's sandbox that analyzes AI-agent skills and prompts for malicious behavior before they reach customer environments, and plans to fold identity threat detection and posture management into one offering while keeping its open integration ecosystem.
▸ The MSP Angle
Do I need identity security that covers AI agents, not just people?
The identity vendors your stack depends on are racing to cover non-human and AI-agent accounts because that is the fastest-growing category of identities in client environments, and an agent with stolen credentials moves faster than any human intruder. This consolidation means agent-aware identity monitoring is becoming a standard layer, not a premium add-on. Get ahead of it: inventory the service accounts and agent identities in each client tenant now, because you cannot monitor what you have not counted.
Read at MSSP Alert ↗
CRN · Jul 29
ServiceNow aims at the Fortune 500,000 with a ticketless, AI-native service desk
On its earnings call, ServiceNow CEO Bill McDermott previewed a product-led offering aimed at small and midsize companies, described as a conversational service desk with no tickets and AI-coded automation, already in beta and near general availability. Partners told CRN that AI-native deployment finally makes the midmarket viable after the failed ServiceNow Express era, with change management rather than configuration time now the limiting factor.
▸ The MSP Angle
Is an enterprise ITSM vendor about to compete for my SMB service desk?
An enterprise ITSM giant marching downmarket with an agentic, ticketless service desk is both a competitive threat and a validation of the AI-service-desk market MSPs already live in. The defensible ground is the part the platform cannot ship: the change management, the client relationship, and the judgment about which work to automate. If you sell managed service desk, get your own AI-assisted intake and resolution story in front of clients before a product-led competitor frames the category for them.
Read at CRN ↗
CRN · Jul 28
Monday.com cuts 20% of staff in a pivot to an AI work platform, and insists AI is not the reason
A July 22 SEC filing shows Monday.com cutting about 20 percent of its workforce, roughly 620 people, with 45 to 55 million dollars in net charges, as it reorganizes over nine months from managing work to doing the work with humans and AI agents in one workspace. Co-CEO Eran Zinman denied the cuts are AI replacement or cost-driven and said most savings will be reinvested, with hiring continuing in strategic areas.
▸ The MSP Angle
Are companies really restructuring their whole workforce around AI agents?
A mainstream SaaS vendor reorganizing its entire company around agentic AI is the exact pattern clients will ask you to interpret, and the messaging fight matters as much as the layoffs. Whether or not you believe the not-AI-replacement framing, the operational reality is that vendors are rebuilding around agents doing work, which raises the governance stakes for every automation a client adopts. Position yourself as the advisor who separates the AI hype from the AI plan, because the client watching this headline wants a translator, not another vendor.
Read at CRN ↗
Domino Data Lab · Jul 21
Domino survey: AI ROI still fails to outpace spend for 57% of enterprises, unchanged since 2025
The fifth annual Domino enterprise AI report, based on 639 director-level AI leaders at firms above 100 million dollars in revenue and fielded by BARC in April, finds 57 percent say AI returns still do not exceed spend, flat across two years, even as 93 percent report improved production capability. Agentic AI is outrunning controls: 43 percent have it in governed production, but 41 percent are piloting or scaling it with no governance in place.
▸ The MSP Angle
Why is AI not paying off for most companies yet?
The ROI number has not moved in two years while production capability keeps climbing, which means the gap is not the technology, it is governance and the last mile to a business user acting on what the model found. That gap is the clearest MSP advisory opening of the year: clients do not need more AI, they need someone to make it pay and make it safe. The 41 percent running agentic AI with no governance are not a warning, they are a prospect list.
Read at Domino Data Lab ↗
Google Workspace Updates · Jul 28
Gemini image and diagram generation arrives inside Google Docs, gated by an admin setting
Google began a gradual rollout on July 28, up to 15 days, letting Docs users generate and edit images, diagrams, and infographics from natural-language prompts on the web. It is available on Business Standard and Plus, Enterprise Standard and Plus, Education Plus, and Google AI Pro and Ultra, and is gated by the Gemini for Workspace in Drive admin setting plus user smart-features toggles.
▸ The MSP Angle
What new Gemini capabilities land in my clients' Workspace tenants this week?
This is an obligation before it is a feature: a new generative capability appears in managed tenants within about two weeks, controlled by the Gemini for Workspace in Drive admin toggle, so the governance question lands before any user asks. Decide the default per client now rather than discovering it live, confirm the admin setting matches each client's data-handling posture, and update the Workspace governance baseline so new Gemini toggles are a documented decision instead of a surprise.
Read at Google Workspace Updates ↗
The Hacker News · Jul 31 
Troutman Pepper · Jul 31