How to pick between Salesforce Agentforce, Microsoft Copilot Studio, Google Agentspace, and a custom agent runtime for your enterprise agent programme.
Artificial Intelligence Solutions
Looking for a artificial intelligence partner?
We build domain-led systems tailored to your industry and workflow. 12 years. 2,100+ engagements.
Every enterprise buyer looking at agent platforms right now is sitting through the same 3 pitches, one after the other. Salesforce is selling Agentforce as the agent layer for CRM-centric work. Microsoft is selling Copilot Studio as the agent layer for anyone already on Microsoft 365.
Google is selling Agentspace as the agent layer for anyone on Workspace and Vertex. Each pitch positions its platform as the obvious enterprise-grade answer. Each contract is worth a lot of money and locks the enterprise into that vendor's agent runtime for years.
Behind the pitches, the honest question is not "which platform is best" but "which platform fits where our data and workflows actually live, and are we sure we want to bet the entire agent programme on any single vendor". Answering that question well is worth more than any of the demo experiences the vendors are running right now.
So how do you pick between them, and when is custom actually the right answer? That is the point of this piece. You will see what these 4 platform options really are without the trademark slides, how they compare on the questions your enterprise actually cares about, the 3 decisions that pick the right platform for your specific situation, the 4 enterprise-specific ways these platforms break in production, a simple pattern that lets your enterprise run agents on any of the 3 platforms plus custom without betting the whole agent programme on one vendor, and the 3 signs the platform being pitched is ahead of what it can actually deliver today. All of it is written for the enterprise buyer signing the multi-year contract, not the vendor demo engineer, because the buyer is the one who will live with the platform choice long after the contract is signed.
Why does this matter more this year than last? Because all 3 hyperscaler agent platforms have moved from announcement to selling-hard mode, and enterprise procurement cycles are turning into commitment cycles. The platform picked in the next few months will shape which agents get built, where the data flows, which team owns the runtime, and what the exit story looks like when the vendor's roadmap and your business's roadmap diverge. Every enterprise that committed to one of these platforms without the sort discovered specific limits after signing; every enterprise that took the time to sort correctly ended up with a platform mix (or an escape hatch) that survives the vendor's next pricing shift.
The rest of this piece walks the answer in the order the questions come up during a real enterprise agent-platform evaluation. What is each platform actually? How do they compare?
Which one fits your enterprise? What breaks after commitment? How do you build so a platform switch is bounded work rather than a rewrite?
And how do you tell when the pitch is ahead of the product? Boring on purpose, because enterprise platform decisions are boring until the moment the vendor's roadmap changes and the boring analysis becomes the reason your enterprise still has options.
80%
Enterprises that will use generative AI APIs or deploy GenAI‑enabled apps by 2026, up from under 5% in 2023 (Gartner, 2023)
65%
Organizations reporting regular use of generative AI across their operations, up from one‑third the year prior (McKinsey, 2024)
23.8%
Salesforce share of the global CRM market by revenue, indicating the largest CRM‑centric platform footprint (Statista, 2023)
Enterprise Agent Platforms, Explained Without the Trademark Slides
What is an enterprise agent platform, in practical terms? An enterprise agent platform is a vendor-hosted runtime for building, deploying, and running AI agents that operate inside that vendor's ecosystem. The platform provides the agent building surface (usually a low-code builder plus code-level customisation), the runtime that executes the agents, the integration surface with the vendor's own applications, and the governance and observability the enterprise team uses to manage what the agents are doing. What is bundled and what is separate varies per vendor.
What is Salesforce Agentforce? Agentforce is Salesforce's agent platform, built to run agents inside the Salesforce data cloud and across the Salesforce customer, sales, service, and marketing applications. Its centre of gravity is CRM-shaped work: sales-cycle agents, service-response agents, marketing-orchestration agents. The tighter your enterprise is wired into Salesforce today, the more of your agent work Agentforce can absorb without additional integration.
What is Microsoft Copilot Studio? Copilot Studio is Microsoft's agent-building platform, tightly integrated with Microsoft 365 (Teams, Outlook, SharePoint, Word, Excel), Dynamics 365, Power Platform, and Azure. Its centre of gravity is the productivity and collaboration layer your workforce already lives in. Enterprises heavily on Microsoft find that Copilot Studio drops into existing workflows with less integration cost than either of the other platforms, and reaches employee productivity use cases the CRM-focused platform is not built for.
What is Google Agentspace? Agentspace is Google's enterprise agent platform, integrated with Google Workspace, Vertex AI, and BigQuery. Its centre of gravity is the data-plus-productivity story: agents that reason over data in BigQuery, act through Workspace apps, and use Vertex AI's model catalogue underneath. Fits enterprises on Google Workspace or with significant data platforms in Google Cloud; less compelling for enterprises whose data and productivity ecosystems live elsewhere.
What does custom mean in this context? Custom is building the agent runtime yourself, on top of the AI orchestration framework of your choice, integrated with the specific systems your agents need to reach. More engineering work up front, more control forever, no per-seat platform tax at scale, and no single-vendor dependence. Fits enterprises whose workflows do not sit inside one vendor's ecosystem cleanly, or whose scale makes the platform's per-seat pricing untenable.
The Platform Question
If you cannot say in one sentence why your enterprise's specific data and workflows fit one platform over the others, you are being pitched by the vendor whose account team reached you first, not by the platform that actually fits your situation. Every honest platform choice starts with where your data and workflows already live, not with which platform's brand is loudest this quarter.
Which of the 4 Enterprise Agent Platforms Fits Your Situation?
Which of the 4 does your enterprise actually need? Almost every "which agent platform" conversation resolves into one of them once you push on where your data lives, where your workforce works, and what your agent use cases look like. Knowing which one you are on changes the integration effort, the per-seat cost, the governance model, and how portable your agents will be if the platform's roadmap drifts.
The 4 Platforms
Which Enterprise Agent Platform Fits Your Situation
1
Platform 1
Salesforce Agentforce
Agent platform inside the Salesforce data cloud, tuned for CRM-centric workflows. Best when the customer, sales, and service data your agents reason over already lives in Salesforce, and when the workflows are within the Salesforce family of apps. Deep integration; tight vendor lock-in.
2
Platform 2
Microsoft Copilot Studio
Agent-building platform woven into Microsoft 365, Dynamics, Power Platform, and Azure. Best when your workforce lives in Microsoft productivity apps and your data spans Dynamics and Azure. Broadest reach across employee productivity use cases; also the widest surface for lock-in creep.
3
Platform 3
Google Agentspace
Agent platform integrated with Google Workspace, Vertex AI, and BigQuery. Best when your enterprise runs on Google Workspace, keeps meaningful data in BigQuery, or already uses Vertex AI models. Strong on data-heavy reasoning workflows; smaller enterprise-application footprint than the other 2 hyperscalers.
4
Platform 4
Custom Agent Runtime
Your own agent runtime on top of the orchestration framework and models you pick, integrated with whatever systems your agents need to reach. More engineering work up front; no per-seat platform tax; zero single-vendor lock-in. Fits enterprises with workflows or data that do not sit inside a single vendor's ecosystem.
Which Enterprise Platform Fits
Salesforce fits CRM-heavy agent work. Microsoft fits enterprises deep in Microsoft 365 and Dynamics. Google fits enterprises on Workspace or with significant data in BigQuery. Custom fits enterprises whose agents need to reach across vendor boundaries or whose scale makes per-seat pricing untenable. Most large enterprises end up running some agents on 2 of these plus custom for the workflows that do not fit any hyperscaler cleanly.
Why does the platform choice matter so much before you commit? Because each platform assumes a specific data-and-workflow gravity: Agentforce assumes your work happens in Salesforce, Copilot Studio assumes your work happens in Microsoft's productivity surfaces, Agentspace assumes your data lives in Google Cloud. Picking the platform whose assumptions do not match your enterprise's actual gravity means fighting the platform's integration model on every use case, which usually looks like "we cannot reach that data cleanly" or "we cannot embed the agent inside the tool the user actually works in". The platform's fit to your gravity is the leverage; the platform's brand recognition is not.
3 Decisions That Pick the Right Platform
Once you know the 4 options exist, which questions actually separate the right one from the noisiest? The 3 decisions below are the ones that keep showing up in enterprise procurement conversations. Every other input (which vendor your CIO's college friend is at, which platform an industry analyst rated highest last quarter) is downstream of these 3.
01
Where Do Your Data and System of Record Already Sit?
Enterprise agents need to reach real business data to be useful. If your customer data is in Salesforce, your operational data is in Dynamics, your product data is in BigQuery, and your project data is in a fourth vendor's tool, no single platform reaches all of it cleanly. The platform whose native data is closest to what your agents will reason over is the platform that pays back fastest. If your data is genuinely split across ecosystems, custom becomes a much stronger option because the platforms all penalise cross-ecosystem reach the same way. Answer this by inventory of actual data locations, not by which vendor your team wishes was the anchor.
02
Which Productivity Ecosystem Does Your Workforce Actually Live In?
Agents your workforce interacts with feel much better when they show up inside the tools your workforce already uses. Microsoft 365 workforces get the smoothest experience from Copilot Studio agents that live in Teams and Outlook. Google Workspace workforces get the smoothest experience from Agentspace agents inside Docs, Sheets, and Chat. Salesforce power-user workforces get it from Agentforce inside the CRM surface. Picking a platform whose productivity ecosystem does not match your workforce means the agents live in a separate surface your users have to switch to, which crushes adoption. Match the platform to the daily-tools reality, not to the CIO's strategic aspiration.
03
How Much Vendor Lock-In Will Your Finance Team Actually Accept?
The 3 hyperscaler platforms produce different lock-in profiles. Agentforce ties you deeper into Salesforce; Copilot Studio ties you deeper into Microsoft; Agentspace ties you deeper into Google. All 3 use per-seat or consumption pricing that scales with adoption in ways that are hard to control mid-contract. Your finance team's tolerance for that lock-in shapes which platform is acceptable, and it shapes whether custom becomes the honest answer for the workflows where lock-in would be strategically expensive. Enterprises with active multi-cloud policies usually end up mixing platforms plus custom; enterprises comfortable with single-vendor commitment lean into their strongest existing vendor.
Answer Before the Vendor Locks You In
Answer the data question first, the productivity-ecosystem question second, the lock-in tolerance question third. The 3 answers together usually name the platform without much argument. Teams that let the vendor drive the sequence often accept lock-in they would have negotiated harder against if they had answered themselves first. Order and independence matter here more than in most enterprise decisions.
4 Enterprise-Specific Ways These Platforms Break in Production
What actually goes wrong with enterprise agent platforms once they are running at scale? The 4 below are the enterprise-specific failure modes that keep showing up. Each looks fine at pilot scale and lands as a real problem at deployment scale, which is exactly when reversing the commitment is expensive. Recognising them at the commitment stage is what saves the finance team a call later.
01
Pricing Spikes When the Agent Population Grows
All 3 hyperscaler platforms price on some combination of per-seat, per-agent, and per-consumption. At pilot scale the numbers look modest; at deployment scale, when a large portion of your workforce is using agents daily, the pricing compounds fast. Vendors know this pattern and their pricing is built to capture that growth. Model the platform's cost at the full-deployment scenario before signing, not at the pilot. If the maths breaks at scale, the platform is a pilot tool only, and the real deployment needs to be on a different platform or a custom runtime.
02
Zero Agent Portability Between Platforms
An agent built on Agentforce cannot be moved to Copilot Studio, and vice versa. The prompts might carry, the tools do not, the integrations definitely do not. Every platform commitment is a rebuild if you ever want to move. This is not a bug; it is how enterprise platforms make their money. Understand the switching cost before you commit, and consider whether the wrapper pattern below is worth building specifically so future migrations become bounded rather than existential.
03
Capability Gap Between Marketing and Real Workloads
The platform's demos are curated to a small set of showcase workflows. Your actual enterprise workflows include edge cases, custom fields, non-standard integrations, and years of accumulated data-model changes. The gap between the platform's demo behaviour and its production behaviour on your real setup is where enterprise projects surprise their sponsors. Insist on a pilot against your real data model and real integrations before signing, not against the vendor's clean demo tenancy. The gap is real, and running the pilot on real data is the only way to size it.
04
Permission Model Mismatches With Your Enterprise IAM
Each platform inherits the permission model of its vendor's ecosystem: Salesforce sharing rules for Agentforce, Microsoft Entra roles for Copilot Studio, Google Workspace and IAM for Agentspace. Your enterprise's actual identity and access management story usually spans multiple vendors and does not map cleanly to any single platform's model. Agents end up with permissions the enterprise IAM would not have granted, or blocked from data your business rules do allow. Reconciling these mismatches is real integration work that vendors rarely mention in the sales conversation.
Why Enterprise Agent Platforms Break Differently
All 4 of these failure modes are enterprise-scale specific: they do not appear in pilot demos and do not appear in the vendor's case studies. They appear the quarter after deployment lands at scale, which is exactly when the multi-year contract is already signed. Model each of them explicitly before committing, and use the results to negotiate specific contract terms (usage caps, migration assistance, IAM extension work) that address them.
Platform Fit Grid
Which Platform Suits Which Enterprise Situation
Situation
Agentforce
Copilot Studio
Agentspace
Custom
Salesforce-heavy
Best fit
Weak
Weak
Fine
Microsoft 365 & Dynamics
Weak
Best fit
Weak
Fine
Google Workspace & GCP
Weak
Weak
Best fit
Fine
Multi-cloud or mixed
Awkward
Awkward
Awkward
Best fit
Strict data residency
Region-limited
Region-limited
Region-limited
You control it
Platform Follows Existing Investment
The right platform is usually the one that matches your existing enterprise investment. Custom fits multi-cloud, strict-residency, or genuinely differentiated workflows where the platform vendors offer generic solutions.
A Pattern That Runs Enterprise Agents Without Betting on a Single Platform
So what does an enterprise agent programme look like when it is built to survive the platform landscape shifting under it? Not fancy. The shape below is the arrangement that lets your enterprise run agents on Agentforce for some workflows, Copilot Studio for others, Agentspace for a third set, and custom for the workflows that fit none of them, all through a single governance layer your enterprise owns.
Every layer has one job. When any single platform's roadmap or pricing shifts, the change stays contained to that platform's adapter.
Architecture
A Pattern That Runs Enterprise Agents Across Platforms Plus Custom
Layer 1
Enterprise Agent Registry
Central catalogue of every agent your enterprise runs, regardless of which platform it runs on. Owners, purposes, permissions, cost, incident history, in one place your enterprise controls.
Layer 2
Agent Gateway
Single entry point for your business systems to invoke agents. Routes the request to the right platform's runtime based on which platform hosts that agent, without the calling system needing to know.
Layer 3
Platform Adapters
One adapter per platform: Agentforce, Copilot Studio, Agentspace, custom. Each speaks the enterprise gateway's interface into the specific platform's agent API.
Layer 4
Enterprise Governance
Your IAM, your cost caps, your audit trail, your policy enforcement, applied to every agent regardless of which platform runs it. Rules live once, at your enterprise layer, not per platform.
↓
What This Buys You
Platform Choice Without Multi-Year Lock-In
Platform Per Workflow
Sales agents on Agentforce, employee productivity agents on Copilot Studio, data-heavy agents on Agentspace, edge cases on custom. All coexist under the same governance.
Migrate Cleanly
When a platform's pricing shifts or capabilities plateau, migrate the agents that platform hosts to another adapter without touching the business systems that call the gateway.
Governance Everywhere
Your IAM, cost caps, and audit trail apply uniformly. No platform gets a governance carve-out because your enterprise layer sits above all of them.
Why Platform Independence Compounds
The hyperscaler agent platforms will keep churning: new features, new pricing, new capability, occasional deprecation. Enterprises that hard-wire themselves to one platform ride each shift as a project. Enterprises with the gateway pattern absorb the shifts at the adapter layer. The upfront investment is real and small compared to the migration cost of the alternative.
Why build this pattern even if you plan to use only one platform for now? Because "for now" tends to become "forever" once the switching cost has been unopposed. Enterprise agent platforms are a category where the ecosystem will keep changing, and the enterprises that keep their options open through the gateway pattern get to make purchasing decisions from strength rather than accepting whatever the incumbent vendor wants next year. The upfront work is small; the negotiation leverage it produces is real for every subsequent contract cycle.
3 Signs You Are Being Sold an Agent Platform Ahead of Its Delivery
How do you tell whether the platform being sold to you is production-ready for your enterprise workflows or still catching up to its own marketing? The 3 signs below are the ones that keep separating real platform capability from launch-cycle enthusiasm. If you spot more than one, the platform is probably closer to the roadmap than the release.
01
All the Reference Agents Come From the Vendor's Own Case-Study Programme
Ask the vendor for 3 enterprises running similar-shape agents to what you want to build, in production, on the specific platform. Not launch-day case studies from the vendor's marketing team. If the answer is a list you cannot actually reach or the references all come from the vendor's paid case-study programme, the platform is probably still in early adoption for use cases like yours. Real production adoption produces peer customers who will speak to evaluators; launch-cycle marketing does not, because the peer pool is not there yet.
02
The Platform Requires Your Data to Already Live in the Vendor's Cloud
The platform's setup guide starts with "first, replicate your data into our cloud". That is a real requirement, not a suggestion, and the replication cost (in engineering, in dollars, in ongoing sync work) is a real part of the total cost of adopting the platform. Vendors selling agent platforms often understate this because it is enterprise-integration friction that undermines the pilot enthusiasm. Ask exactly what data has to move for your specific use cases to work, and add that cost to the sticker price before comparing.
03
Agent Runtime Cost Is Unclear at Deployment Scale
Ask the vendor to model the platform's cost at 10 times your pilot scale. If the answer is vague, or the vendor asks to schedule a follow-up to "work through the details", the pricing model has not been validated at that scale even by the vendor's team. Platforms with mature pricing model out to enterprise scale confidently; platforms with new pricing offer flexibility that is really uncertainty. Getting the scaled-out pricing on paper before signing is the difference between a predictable enterprise expense and a surprise bill quarter after next.
The Platform-Delivery Filter
Ask the vendor 3 things in the same meeting: introduce me to 3 reachable customers running similar agents in production; itemise every data-movement and integration cost my use cases require; and model your pricing at 10 times my pilot scale on paper. Platforms ready for enterprise-scale deployment answer all 3. Platforms still catching up answer with adjectives, deferrals, or "our team will follow up".
Frequently Asked Questions
Should you standardise on one enterprise agent platform, or run a mix?
For large enterprises with agents across multiple ecosystems, a mix is almost always the honest answer. Different workflows sit on different platforms cleanly, and forcing everything into one platform means fighting the platform's gravity for the workflows that do not naturally fit it. Standardising on one platform makes sense for smaller enterprises heavily concentrated in one vendor's ecosystem. The gateway pattern above lets larger enterprises mix cleanly without giving up governance, which is what makes the mix workable rather than chaotic.
When is custom really the right call versus adopting one of the 3 platforms?
Custom fits when your agent workflows cross vendor boundaries the platforms cannot bridge cleanly, when your scale makes per-seat pricing untenable, when your governance model is strict enough that no vendor's platform-native model meets it, or when your differentiation depends on agents doing things no platform is built to support. For any of those, custom is the honest answer. For workflows that fit inside one vendor's ecosystem, run at moderate scale, and do not need unusual governance, the vendor platform is usually faster and cheaper than building custom for that specific workflow.
How does the enterprise gateway pattern actually reduce lock-in?
By isolating platform-specific code inside the adapter layer instead of scattering it through the business systems that call agents. Without the gateway, your CRM code calls Agentforce directly, your productivity tool calls Copilot Studio directly, and moving any agent means rewriting every calling system. With the gateway, your business systems call the gateway with an agent intent, and the gateway routes to whichever platform hosts that agent. Migrating an agent from one platform to another becomes an adapter update plus a registry change, not a rewrite. The lock-in shifts from your business systems to the adapter layer, which is much cheaper to change.
Do these platforms handle compliance requirements out of the box?
Each has its own compliance posture matching its vendor's broader enterprise programme. None of them handle every regulator's requirements automatically; they provide the platform capabilities you can compose into a compliant deployment, but the deployment work is still yours. For regulated industries (healthcare, finance, government), verify the specific certifications and data-residency options against your specific obligations before committing. Do not assume "enterprise-grade" in the marketing means the platform meets your specific regulator's rules; ask for documentation and read it.
Can you build an agent on the vendor's low-code builder and migrate it to a custom runtime later?
Usually not cleanly. The low-code builder produces agents in the platform's specific format, using the platform's specific tools and integrations. Migrating that to a custom runtime means rebuilding the agent from scratch; the visual configuration does not export. Consider low-code builders as a fast path to production on a specific platform, not as a prototype that can move later. If migration flexibility matters, build the agent through the platform's code-level API rather than the low-code builder, or on custom from the start.
Which platform has the best model catalogue underneath?
All 3 now offer access to multiple models: Agentforce runs on the underlying models Salesforce contracts, Copilot Studio on Azure OpenAI plus other Azure-hosted models, Agentspace on Vertex AI's catalogue including Google's own models and third-party ones. In practice the model catalogue matters less than the platform's fit to your data and workflows, because the model is a smaller part of the total value than the integration and workflow layer. Pick the platform on data-and-workflow fit; verify the model catalogue is acceptable but do not let it be the deciding factor.
Can Entexis help you pick the right enterprise agent platform and build the gateway?
Yes. Entexis designs and builds enterprise agent programmes across Agentforce, Copilot Studio, Agentspace, and custom runtimes, with the enterprise gateway pattern that keeps your programme free from single-vendor lock-in. That work starts with the honest sort against your actual data-and-workflow reality, not against whichever vendor reached you first. We then design the agent registry, wire the gateway with adapters for the platforms you actually pick, and stand up the enterprise governance layer that applies your IAM, cost caps, and audit trail across every agent regardless of platform. Reach out with where your data and workflows sit, which vendors are already in your enterprise, and which use cases are the priority, and we can walk through the right platform mix for your specific situation.
So where does that leave your enterprise agent platform decision? The 4 options above are the real choices every enterprise buyer is evaluating today. Agentforce fits CRM-centric agent work when Salesforce is your data anchor.
Copilot Studio fits enterprises deep in Microsoft's productivity ecosystem. Agentspace fits Google Workspace enterprises and data-heavy reasoning workflows on BigQuery. Custom fits workflows that cross vendor boundaries or scale past per-seat pricing tolerance.
The 3 decisions above pick between them per workflow; the 4 enterprise breakers are what surprise buyers who signed at pilot scale; the gateway pattern lets your enterprise mix platforms cleanly and keeps future migration bounded. Sort honestly by where your data and workforce already live, negotiate contracts with the scaled-out failure modes in mind, and build the gateway from day one so no vendor's roadmap ever holds your entire agent programme captive.
Want to Pick the Right Enterprise Agent Platform Instead of the Loudest Contract?
At Entexis, we design and build enterprise agent programmes across Agentforce, Copilot Studio, Agentspace, and custom runtimes, unified through the gateway pattern that keeps your programme free from single-vendor lock-in. We start with the honest sort against your actual data-and-workflow reality, negotiate against the enterprise-scale failure modes before they land as surprise bills, wire the gateway with adapters for the platforms you actually pick, and stand up the enterprise governance layer that applies your IAM and cost caps across every agent. Your enterprise gets to pick the right platform per workflow, keep the option to change, and negotiate every future vendor contract from strength. Start the conversation with Entexis.
Ready to Add AI to Your Business?
From intelligent chatbots to workflow automation, we build AI solutions that understand your domain, your data, and your users. Tell us what you need.
We'll get back within one business day.
Thank You!
We've received your message and will get back to you within one business day.
Try the AI workflows we build, for real, right now.
Same workflow patterns Entexis rolls into client setups. Try them in your browser, no signup. If one feels like it'd help your team, we build a private version tuned to your data.
