In April 2026, Skai did something quietly revealing for an enterprise suite: it launched an MCP interface, letting customers run their own AI agents against its unified cross-channel data without custom API work. A month later it announced Skai Studio, an agent-native environment where teams deploy specialized agents that detect performance shifts, diagnose root causes, adjust budgets, and notify stakeholders — with human approval checkpoints wherever the organization wants them. When a platform's newest surfaces are built for agents, the direction of travel is not in dispute: the reasoning is moving out of the console.
Which sets up this page's question precisely. If agents are doing the work either way, what does the platform itself consist of — and which of its parts could you assemble directly, with Claude on MCP servers for the agent loop, Cursor for the scripts and tools, n8n and BigQuery for the plumbing?
Verified July 2026 against Skai's published announcements; re-check before relying on any capability claim.
What Skai is, and what it's made of
The credentials are earned. Skai — formerly Kenshoo, renamed in 2021 after acquiring Signals Analytics — is an omnichannel platform for commerce media spanning retail media, paid search, and paid social. It cites integrations with more than 300 publishers and retail media networks, Amazon Ads, Walmart Connect, Criteo, Google, Microsoft, Meta, and TikTok among them, and more than 8,000 brands and agencies as customers, including PepsiCo, Estée Lauder, Publicis, and WPP. Its AI layer — Celeste, a generative agent for commerce media, plus the Studio environment above — is a real bet, not a badge.
Structurally, though, the platform decomposes the way every entry in this cluster does: connectors pulling performance data from publishers and retail networks; decision logic turning data into proposed or automated changes; execution calls pushed back through the same publisher APIs any advertiser can access; and a console over the top. Keeping hundreds of connectors alive as publishers change their APIs is hard, unglamorous work, and a coherent interface over that sprawl has genuine enterprise value. But none of it is privileged infrastructure — and Skai's own MCP move concedes that the layer above it, the reasoning, is now portable.
The migration Skai's roadmap confirms
This shape of events is the thesis we keep returning to: modernizing an entire ecosystem is easier than doing it one piece at a time — and the modernization underway right now is business logic leaving structured systems for intelligent agents, while the systems themselves settle into the role of ledgers and connectors.
A commerce-media suite shipping an agent interface is that thesis playing out on schedule: the platform repositioning itself as ledger and connector layer while the reasoning moves outside. Which leaves one question standing — whether the agents doing your reasoning belong to the vendor or to you. (The cluster-wide version of this argument is in our build-vs-buy framework piece.)
Layer by layer: the direct build
| Layer | Inside Skai | The direct build |
|---|---|---|
| Data connectors | 300+ publisher and retail media network integrations, vendor-maintained | MCP servers plus scheduled exports into Sheets or BigQuery, for the channels you actually run — data pipeline integration |
| Decision logic | Celeste and Studio agents that detect shifts, diagnose causes, and recommend or adjust | Claude reasoning over your data with rules in plain language; Cursor hardens the stable parts into scripts — PPC intelligence |
| Execution | Changes pushed through publisher APIs inside organization-defined checkpoints | The same publisher APIs via MCP write actions or n8n, under spend caps you set — automation governance |
| Interface | A unified enterprise console for planning, activation, and measurement | A dashboard your team actually reads, plus the agent conversation itself — data access & dashboards |
Two clarifications keep the table honest. First: you will not rebuild 300 connectors, and shouldn't try — a direct build covers the two to five channels where your spend actually lives, which for most teams outside retail-media-at-enterprise-scale is the entire problem. Second: a build is not a one-shot prompt. Each guide in the automations library is a working system with a prove-it-on-exports stage before anything connects live, and the setup guide puts access, accounts, and guardrails in place before anything touches spend.
Where buying keeps winning
- Breadth is your actual problem. Coordinating retail media across dozens of networks is what Skai's connector coverage and console are for. Rebuilding that footprint is not a rational project.
- Many operators, one tool. Twenty people across three agencies need permissions, views, and workflows an internal build won't match soon.
- Day-one value and a phone number. A platform works when you sign; a build works when you've built it.
- Nobody will own it. Owning even a small, well-documented system is a commitment some teams rightly decline.
And where building wins: decision logic that carries your margin structure, seasonality, inventory constraints, and brand rules directly instead of through configuration options; performance history in your own warehouse, queryable for whatever you think of next; running costs instead of vendor-priced (and vendor-re-priced) licensing; a feature pipeline that is a build sprint rather than a request queue; and guardrails — caps, ceilings, exclusions, thresholds — that are yours to set and audit.
That trade always existed. What moved is the price of the second column — and the reason it moved is worth one more section.
Every term in the old equation inverted
The twenty-year doctrine of enterprise software — Salesforce its best-known evangelist — held that customization was the risk: expensive to write, brutal to keep, hostile to upgrades. Fit your business to the software, because the reverse was unaffordable.
AI inverted each term. Writing the adaptation: hours, not quarters. Changing it: a revised plain-language rule, an afternoon. Maintaining it: the same tools that wrote the code read it, explain it, and repair it when an API shifts. Once customization stops being a liability, out-of-the-box workflow stops being an asset — it just means running the same logic as every other holder of the same subscription. The unique fit is the differentiator now, and for the first time it is cheap to have.
The path, if you build
The automations library is the concrete version of this page: build guides for the workflows platforms are typically bought for — search-term intelligence, budget reallocation under caps, measurement repair, competitive monitoring — each with the exports-first route and the MCP route, on top of one shared setup. Use the guides solo, or build with us: the Automated Campaign Optimization Campaign Automation Audit is the first sprint, and the sprints after it each deliver one governed workflow, every change logged as Trigger / Action / Impact on accounts you own.
Get the build-vs-buy call right before committing either way
The decision gets easier once you know what your accounts, data, and team can support. The free Readiness Score — four minutes, no login required — tells you whether to start with a guide, an audit, or a platform.
Get your free Readiness Score →Keep reading
- campaignautomation.ai vs. Skai — the direct head-to-head if you're weighing the platform against a governed build.
- Skai vs. Marin — the two enterprise cross-channel suites compared.
- Skai vs. the governed-automation approach — the control model behind everything in the library.
- Build vs. buy in AI marketing automation — the anchor essay for this whole cluster.