AI Stopped Selling
Software.
It Started Selling Outcomes.
Salesforce now charges per resolution. Anthropic and Blackstone put $1.5B behind implementation as the product. The week the pricing model — not the model — became the story.
Pay-Per-Resolution Is Live.
Your Data Layer Just Became a P&L Line.
Salesforce's Agentforce Help Agent went generally available this month with a pricing model that would have sounded reckless two years ago: charge only when the agent resolves the case. Deployment spans all channels, and the vendor absorbs the risk of the agent failing. The m3ter acquisition — closed July 1 — brings the metering and rating infrastructure to bill this way at enterprise volume.
Read the incentive structure carefully. Outcome pricing only works for the vendor when resolution rates are high, and resolution rates are a function of data quality, not model quality. An agent grounded in a fragmented customer record deflects tickets. An agent grounded in unified, governed data resolves them. The vendors know this — which is why the same release cycle pushed Data 360 and catalog integrations so hard.
"Before signing an outcome-priced agent contract, audit whether your resolution events are even instrumented. If you cannot independently verify what 'resolved' means in your systems, you are negotiating blind against a vendor who can define it for you."
The billing model for all of them is being decided in procurement conversations happening right now. The projects that survive are the ones where the data foundation, the governance model, and the application layer were designed as one system — not assembled separately and integrated at the end.
The SI Land Grab: Delivery Is the New Battleground.
Four Shifts Deciding Who Wins It.
Three global system integrators announced new Anthropic alliances between July 10 and July 15 alone — each standing up a "Claude Center of Excellence" pattern: reusable agent skills, reference architectures, and a governance layer sold as a managed service. The Anthropic-Blackstone $1.5B implementation venture confirms the thesis the labs have now acted on: distribution through delivery partners — not direct enterprise sales — is how agentic AI actually reaches the enterprise at scale.
"Winning enterprise customers requires far more than shipping better models."
Here is what the press releases skip: a partnership announcement is not a delivery capability. The firms winning agentic engagements already own the data layer underneath the agent — because that is where deployments consistently stall. Generalist alliances built on top of a broken data pipeline produce a faster path to a canceled project, not a working one.
Catalogs Inside ChatGPT. Governance Inside the Agent's Execution Path.
Two Platforms, One Direction.
Salesforce is putting commerce data in front of AI assistants it doesn't control. Snowflake is putting guardrails inside the path where agents actually act. Different problems, same fortnight, same underlying dependency on data quality.
Mayfield's survey of 266 enterprise technology leaders surfaced a quiet power shift: agents are being procured the way SaaS was in 2012 — department by department, ahead of central governance. The gap between adoption and survival is architecture.
46% of purchases now come from LOB leaders, ahead of CIOs and CTOs at 38% each. Agents are being procured the way SaaS was in 2012 — department by department, ahead of central governance.
Roughly 42% of enterprises report agentic AI in production, with 72% in production or pilots combined . The deployment numbers explain the urgency on both sides of the buying decision.
On cost, unclear value, and inadequate risk controls. The delta between the 72% in pilot/production and this cancellation forecast is architecture — projects assembled separately and integrated at the end don't survive.
Business units choose the agent, IT inherits the data problem, and the central architecture conversation happens after the deployment is already running on customer data. The organizations getting ahead of this run the data readiness audit before the agent RFP is issued.
When software does the buying, someone still has to own the trust. This episode maps the three competing architectures for agentic commerce: the open ecosystem forming around Google, Anthropic, OpenAI, Stripe, and Shopify's competing protocols; the integrated ecosystem Amazon and China's Alibaba, Tencent, and ByteDance are building by keeping the entire purchase journey in-house; and the open network model India is scaling through ONDC and UPI, now extending to agent identity and agent payments.
The episode closes on the six trust checkpoints — recognition, discovery, evaluation, payment, fulfillment, accountability — that decide whether an agent ever recommends your business at all. Which checkpoints you control, and which you cede to the platform, is a strategic decision with commercial consequences most architecture reviews haven't reached yet.
Running Agents in Production?
That's the Conversation We're Building.
The conversations in this episode — on who owns trust in agentic commerce, which architectural decisions lock you into a platform, and what the six trust checkpoints cost you in dependency — are the kind TeqTalk is built around. Practitioners with production deployments and real specifics. CIOs, CDOs, enterprise architects. No vendor pitches. No vision decks.
If you have taken agents to production, built the data layer underneath them, or navigated the governance conversation after deployment — we want to have that conversation.
Apply to Be a Guest →