Charting our Future: Why APPLY is Betting ACx on Google Cloud

What this is not

Let me be clear about what this decision is not. It is not a wall. Our clients run on AWS, Azure, Databricks, and Snowflake, and we will meet them where they are with the same standard of care. But meeting a client where they are is different from where we place our bets. Our investment, our certifications, our reusable assets, and our roadmap now point in one direction. It is not a claim that Gemini is always the best model for every task; Claude and GPT will be the right answer in many cases, and Google's own platform makes that easy. And it is not a bet on a vendor's logo.

It is a bet on strategy. I believe the winners of the agentic era will be the companies, and the partners, who treat models, agents, cloud, and data as one system rather than four purchases. Google is the only one who has built that system end to end. It is also, notably, the only one that has chosen to open the system rather than wall it in.

The four layers of an agentic business

When a brand puts an agent in front of a user, whether it is a shopping concierge, a service rep, a fan-identity engine, or a factory-floor assistant, four things must work together.

First, a model that can reason, see, hear, and speak. Second, an agent layer that gives the model tools, memory, identity, context, guardrails, and a way to act. Third, a cloud that runs all of it at a cost, latency, and security posture an enterprise can stand behind. Fourth, data that gives the agent something true to say about this customer, this product, this machine, this moment.

Pull any one out and the experience degrades in a predictable way. A brilliant model with no data hallucinates confidently. Rich data with no agent layer produces dashboards instead of actions. An agent with no governed cloud never leaves the pilot. And a cloud with neither frontier models nor a coherent agent story is just very expensive plumbing.

Here is how I see the field, and I will be opinionated about it because our clients pay us for judgment.

Anthropic and OpenAI: extraordinary at the top of the stack

Anthropic and OpenAI build the most capable models and increasingly the agents around them, and Claude and GPT have changed how our people work, but model leadership rotates and is measured in months. Neither company is trying to be the place where your enterprise data lives, where access policies are enforced, or where workloads run at global scale. For a client, their limitation is not intelligence but that intelligence alone is not an operating environment.

AWS and Azure: the clouds, with the model story on loan

AWS and Azure are formidable, with enormous enterprise share and deep data services. But their frontier-model story rests on outside relationships (a model lab for Azure, a marketplace and an investment for AWS), which is not the same as building the model, the chips, the cloud and the agent platform as one engineered system. That integration shows up in economics, latency, security and how fast new capabilities reach customers, because a new Gemini capability lands in the same platform, under the same governance, next to the same data. A stack assembled from several vendors' roadmaps inherits the gaps and disconnects between them.

Databricks and Snowflake: the data gravity, and the question of what comes next

Databricks and Snowflake made the lakehouse and cloud warehouse mainstream, and we will keep working with the many clients who run on them. The question CIOs now ask, though, is how to give an agent trustworthy context across every system without copying data everywhere, and Google's answer at Google Cloud Next '26 was the most coherent I have seen. BigQuery runs processing and AI on the same data, a Knowledge Catalog maps business meaning and federates with systems like Salesforce, SAP, ServiceNow and Workday without copying, and an open Apache Iceberg lakehouse reaches data in AWS, Azure, Databricks and Snowflake. Because Google makes existing investments reachable by agents rather than asking clients to abandon them, projects with messy, multi-vendor estates ship instead of stalling in a migration debate.

Where Google spans all four

Now put the four layers side by side.

Models: frontier intelligence that sees, hears, and creates

Google's Gemini 3 family is built from the start to be multimodal. It reasons across text, images, audio, and video in one model, rather than stitching separate systems together. That matters because customer experience is not a text problem. A shopper sends a photo of a sofa and asks for a rug that matches. A fan asks about the play they just watched. A technician holds a phone up to a failing part. Each of those is a multimodal request, and a model that handles it natively is faster, cheaper to run, and more accurate than a chain of specialists.

Google also builds the creative models that sit beside Gemini, including image and music generation, so one platform can reason about content and produce it. And it is not a closed shop. Model Garden offers more than 200 models, including Google's open Gemma family and leading third-party models, so a team can choose the right model for each task without leaving the platform, its security controls, or its data.

Agents: the part that decides whether an enterprise says yes

Building a clever agent is easy. Running thousands of them safely inside a large company is hard, and that is the problem Google has gone furthest to solve. Vertex AI has grown into the Gemini Enterprise Agent Platform, and its scope shows what agents need in production.

A development kit, now at version 1.0+ across Python, Go, Java, and TypeScript, lets engineers build multi-agent systems in code. Agent Studio gives teams a path from no-code prototypes to full-code builds. A runtime provides sub-second starts and long-term memory, so an agent remembers a customer's preferences across sessions. A sandbox contains the code an agent runs. Then come the controls a Chief Information Security Officer asks for first: a cryptographic identity for every agent, a registry of approved tools, a gateway that enforces policy, Model Armor protection against prompt injection, real-time anomaly detection, and simulation and evaluation tools that test agents before and after launch.

Interoperability is built in. Google helped create the Agent2Agent (A2A) protocol, now governed by the Linux Foundation, so agents built by different teams and vendors can work together. It also offers managed Model Context Protocol (MCP) servers for services such as BigQuery and Google Maps, so agents can reach data and tools through a governed standard rather than custom integrations.

For ACx in particular, Google ships Gemini Enterprise for Customer Experience, with prebuilt shopping and food-ordering agents, an omnichannel gateway that carries context between web, mobile, and voice, and agent-assist tooling for human teams. These are the building blocks of ACx, available in one governed platform.

Cloud: infrastructure designed for agents, at the center and the edge

Agents change what a cloud has to do. Workloads arrive in bursts, run for long periods, call models constantly, and need to respond in real time. Google's infrastructure is shaped for that. It runs its own TPUs and custom Arm-based Axion processors, operates one of the largest private networks in the world, and has pushed autoscaling and per-second billing into BigQuery and its agent runtimes so that costs follow actual use.

Google Distributed Cloud extends the same stack to places the public cloud cannot reach: a retail store with unreliable connectivity, a factory floor with strict data-residency rules, a ship at sea. An agent that works only when the network is perfect is not a customer experience. It is a demo. Running Gemini and the agent platform at the edge, under the same controls, closes that gap.

Google Cloud is also growing quickly. Its market share is smaller than the two largest clouds, but its growth rate has been among the fastest, and enterprises are increasingly choosing it specifically for AI. For a partner like APPLY, that momentum matters. It means investment in the platform, a growing ecosystem, and clients who are asking for it.

Data: the context that makes an agent trustworthy

An agent is only as good as what it knows, and this is where Google's integration pays off most. BigQuery now runs analytics and AI on the same data. Its SQL functions parse documents, classify text, forecast time series, and run zero-shot predictions on tables without moving data into a separate machine learning system. Hybrid search combines semantic and keyword retrieval for grounded answers, and BigQuery Graph maps the relationships between entities, such as customers, households, products, and parts, so an agent can reason about connections and not only rows.

On top of that sits the Knowledge Catalog, which Google describes as a universal context engine. It captures what the data means to the business and federates, without copying, with systems such as Salesforce, SAP, ServiceNow, and Workday. A cross-cloud lakehouse built on open Apache Iceberg lets agents reach data that lives in other clouds and platforms, with caching that reduces egress costs. Operational databases such as Spanner and AlloyDB add transactional data, with vector search at the scale of billions of vectors, to the same picture.

The principle is simple and correct: move the question to the data, not the data to the question. Clients arrive with messy, multi-system estates. A platform that is open enough to reach all of it, and intelligent enough to make it usable by agents, is the difference between a project that ships and one that stalls in a migration debate.

Each of these is, on its own, competitive with the best specialist. The difference is that they were designed to meet. The agent calls the model, which reads from the data, on the cloud, under one identity and governance plane. Everyone else is, at best, assembling that experience through integrations. For ACx, which lives or dies on the seamlessness between those layers, I would rather start from a stack that was designed as one.

What this means for our clients

At the end of the day, everything we do is in service of our clients. I promised depth in three arenas. These are the ones where APPLY lives, and where I see the evidence most clearly.

Consumer (CPG & Retail): the era of the conversational storefront

Consumer brands are moving from search boxes to concierges. Google's own proof points are instructive. Macy's launched "Ask Macy's," a multimodal shopping assistant that handles text, images, and virtual try-on. Home Depot built agents that capture the expertise of its associates and can identify a caller's need in seconds. Papa John's became the first customer of a unified voice-and-text ordering agent that remembers preferences and past orders. And on the product side, Mars and Unilever are using Gemini Enterprise as an operating layer for their workforces, with Unilever applying multi-agent systems to procurement.

What these share is that the customer-facing agent is only as good as the data behind it: inventory, preferences, loyalty, context. This is exactly where Google's integration pays. A consumer brand can unify first-party data in BigQuery, ground a Gemini-powered agent in it, deploy it across web, app, and voice through one omnichannel layer, and govern it centrally.

We have already seen what this feels like from the inside. Kraft Heinz's TasteMaker platform, one of the projects that earned APPLY our place in the Google Cloud ecosystem, cut concept production from eight weeks to eight hours, using BigQuery, Vertex AI, and Gemini together. That is not a chatbot improvement. That is a different speed of business. It is also a clean illustration of the thesis: the value did not come from one layer. It came from the layers working as one.

For our consumer and CPG clients, retail media, household-level personalization, and agentic commerce are the next frontier, and each demands exactly this combination of rich first-party data, multimodal reasoning, and governed agents.

Entertainment (Sports & Media): identity, content, and the creative loop

Media and sports are about relationships. A fan is not a transaction; a fan is an identity that persists across a stream, a ticket, a jersey, a podcast, and a group chat. The brands that win will be the ones who recognize that person everywhere and respond in kind, without crossing the line from delightful to creepy.

Google is unusually well-positioned here for a reason that is easy to overlook: it is a media company. YouTube, a leading multimodal model family with strong image, video, and audio generation, and the cloud infrastructure behind it all sit within one organization. Gemini's native multimodality, along with Google's generative media models, means an agent can reason about a highlight reel, a script, or a live broadcast as naturally as it reasons about text.

The creative applications are only half the story. The other half is operational: the agents behind the experience that fans never see but always feel. Capcom, one of the world's great game studios, is using Gemini-based agents for visual inspection and playtesting, freeing developers for creative work. Virgin Voyages built "Rovey," a personal concierge agent for its ships, running on Gemini Enterprise and Google Distributed Cloud so it keeps working with limited connectivity at sea. That second detail matters: experience at the edge, where connectivity is imperfect, is where platform depth shows.

For the fan-identity and digital-experience work at the heart of our entertainment and sports practice, the combination is compelling: identity resolution and audience data in BigQuery, a multimodal model to understand content and context, agents to personalize and act, and a cloud that can scale to the moment a championship ends and ten million people open the app at once.

Manufacturing: from dashboards to decisions

Manufacturing is where I believe agentic AI will deliver the clearest, most measurable return, and also where the full-stack argument is strongest. Factories generate staggering volumes of operational data, but the knowledge that makes them run is tacit, stuck in the heads of experienced people and in scattered systems.

The proof points are striking. GE Appliances has deployed more than 800 AI agents across manufacturing, logistics, and supply chain. Tata Steel built over 300 specialized agents in nine months using a low-code platform. Danfoss automated roughly 80 percent of the transactional decisions in email order processing and cut response times from 42 hours to near real time.

Why does Google fit? Because manufacturing needs everything at once. It needs data that spans ERP, MES, supply chain, and sensor streams, which is the Knowledge Catalog's federation story with SAP and others. It needs multimodal models that can read a drawing, inspect a part, and parse a document, which Gemini and BigQuery's document-parsing functions address. It needs the agent layer to be governed, because an agent that touches a production schedule had better have an identity and an audit trail. And it needs infrastructure that can run at the edge, which Google Distributed Cloud provides. Again: four layers, one design.

Why this matters for ACx

APPLY's thesis is that the next generation of customer experience is agentic: systems that understand a customer's intent, hold context across channels, and act on their behalf, within guardrails the brand controls. This is not a front-end problem. It is a full-stack problem, and treating it as a front-end problem is why so many AI pilots impress in a demo and disappoint in production.

A services firm earns its keep in the seams, the places where models meet data, agents meet systems of record, and experiences meet real customers. The fewer seams a platform leaves, the more of our effort goes into the work that is distinctly ours: strategy, design, craft, and change. Google's full-stack architecture does not remove the need for that work. It concentrates it. That is exactly where an agency that has built its identity on human insight and engineering excellence wants to be.

Beyond strategy, this partnership delivers clear practical value. Deep collaboration grants early roadmap visibility, co-investment opportunities, joint go-to-market execution, and direct client access to Google’s engineers, programs, and funding. Rather than diluting our expertise across multiple platforms, our teams achieve deep mastery, earning certifications and building reusable assets that elevate our capabilities.

To accelerate client time-to-value, APPLY developed TORQ™, a suite of proprietary accelerators designed to deliver measurable returns on Google Cloud investments in weeks rather than years:

1. TORQ™ Marketing: Scales content creation and optimizes production workflows.

2. TORQ™ Innovation: Drives rapid product conceptualization and development.

3. TORQ™ Data: Deploys agentic workflows for real-time data management, activation, and measurement.

4. TORQ™ Commerce: Enhances retail readiness, PDP development, and occasion-based shopping experiences.

Focus is the ultimate form of client respect. When we claim expertise on a platform, we mean it.

The decision

Strategy is mostly the art of choosing what not to spread yourself across. We have chosen to make Google Cloud the center of gravity of our agentic practice, because it is where our clients' four layers can finally be one, and where our own craft can matter most.

The brands we serve in consumer, entertainment, and manufacturing do not need another proof of concept. They need an agent that knows them, that acts well, and that they can trust. Building that requires a stack that was designed to be whole.

That is why we chose Google Cloud. And that is why I believe it’s best for ACx and what’s still ahead of us.