Which One Do I Adopt?

I'm Tony Uphoff, and this is the Uphoff on Media podcast, a practical take on how technology is rewriting the way B2B companies get built and run and how to get ahead of it for those founding, running, and funding B2B information, marketing, and technology companies, the businesses that run business. Okay, let's get to work. This episode is titled, Which One Do I Adopt? A path forward for B2B leaders placing their AI bets amid rapid change and vendor uncertainty. Which one do I adopt? That question is landing in my inbox constantly right now. Most B2B leaders have already done the easy part— a pilot here, a chat license there, some low-cost experimentation to see what generative AI could do. What they're stuck on now is the harder question. The one with real budget and real organizational change behind it. Which platform, which vendor, which tools do we actually standardize on? That decision is proving a lot harder to make than the first one, and it's not because leaders are behind. It's because 2 things are moving under them at once. The technology is moving faster than the vocabulary. The first problem is pace, and it hits non-technical leaders and technical ones differently. For the non-technical leader, the terminology itself has become a trap. Generative AI and agentic AI gets used interchangeably in board decks and vendor pitches, and the distinction actually matters. One drafts and summarizes, the other does work with some autonomy over the next step. I've sat in rooms where, yes, we have an AI strategy, turned out to mean a company bought a batch of chat licenses 18 months ago and hasn't touched the question since. That's not a strategy, it's a purchase order that got mistaken for one. For the more technical leader, the problem is different and in some ways worse. They can see the pace clearly enough to be afraid of it. Commit to a platform now and there's a real chance a materially better approach exists within a year, sometimes within a quarter. Remember prompt engineering? Companies were hiring for that role 2 years ago. It's already given way to what people are now calling loop engineering, a fundamentally different way of thinking about how AI systems interact of the workflow over time. That's not a slow evolution. That's a job description going obsolete inside a single budget cycle. FOMO and rational caution are pulling the same leader in opposite directions, and both instincts are correct. Even the vendors don't know what they're selling yet. The second problem isn't the buyer's fault at all. The vendors are still figuring out how to position and package what they're selling, and that uncertainty is landing directly on the customer's desk. 2 default responses show up when a market is this unsettled. Some leaders fall back on the relationship they already have. We're a Microsoft shop, full stop. Decision made by inertia rather than evaluation. Others go the opposite direction and run genuine bake-offs, testing platforms against real workflows before committing to anything. Both are rational responses to vendors who haven't settled on what they're selling. Neither is a strategy on its own. That uncertainty shows up directly in how vendors are going to market. In early May, OpenAI and Anthropic each announced their own enterprise services divisions, investing billions of dollars in forward-deployed engineers who sit inside a customer's team to bridge the gap between the model and the actual application. As Bob Evans, founder of Cloud Wars, has tracked in his analysis of what he calls the AI deployment wars, that's one bet among several playing out at once. Google Cloud made a different one, committing $750 million almost entirely to its partner ecosystem rather than building a deployment arm of its own, and tied a 63% Q1 growth number directly to owning the full stack infrastructure and AI together. 3 vendors, 3 answers to the same question, and no consensus yet on which one wins. Underneath that shift sits a bigger one. Vendors are moving from selling seats to selling outcomes. A per-seat license is easy to price and easy to walk away from. A services engagement built around a specific workflow outcome is neither, and it forces the vendor to prove the technology works before the invoice gets paid, not after. That's becoming a core go-to-market strategy in its own right. Earned trust demonstrated on real workflow is turning into the differentiator that positioning alone can't manufacture. The vendors investing in deployment capacity are, whether they'd frame it this way or not, choosing to compete on proof rather than message. Layer on top of that a plain old Gartner hype cycle dynamic, except this one is running in both directions simultaneously. AI is overhyped in the press and underhyped in daily practice at the same time. Oversold as a magic fix in the pitch deck and still underused in the actual workflow. And underneath all of it sits the honest reason this decision carries more weight than a normal software purchase. Reimagining a workflow is a polite way of saying some of that workflow and the people doing it won't be needed in its current form. Usage is climbing and so is the anxiety that comes with it inside the same company at the same time. That's not a communications problem. It's the real issue leaders are weighing with every platform decision. So how do you actually move forward? Given all of that, waiting for the fog to clear isn't a strategy either. It won't clear on the vendor's timeline, and standing still has its own cost. The better approach is a deliberate staged process built around specific use cases rather than a single enterprise-wide bet. 5 paths cover most of the ground, and each one demands a different kind of discipline. Number 1, embedding intelligence in a customer-facing product. Start narrow and start with a pilot that has a clear success metric before it begins, not after. This is model layer work, and it rewards picking the platform with the strongest track record on the specific capability you need— coding, long-context reasoning, or reliability under agentic autonomy— rather than the broadest brand name. Run it for one product line before you touch a second. Number 2, giving the broader workforce a general assistant. This is the lowest risk, highest adoption use case, and it's exactly where most companies already are. The mistake is stopping here and calling it a strategy. Treat it as the on-ramp with a real measurement plan for adoption and output quality, not just seat count. Number 3, extending a platform relationship you already have. If you're already deep in Microsoft or Google's ecosystem, there's a legitimate case for staying there for lower stakes use cases. The discipline required is separating least resistance from best fit explicitly, use case by use case, instead of letting the existing contract make every decision by default. Number 4, building agentic workflows with a deployment partner. This is the highest stakes, highest payoff use case And it's the one where the vendor's own service investment matters most. Before committing, find a partner, whether that's the vendor's own deployment team or a systems integrator with documented expertise implementing agentic workflows in your specific industry. Ask for named references in your vertical, not general use case studies. Run the pilot on a workflow with a real owner and a defined outcome, not an innovation lab side project. with no one accountable for the result. And number 5, bundling AI into infrastructure you already own. For companies with heavy existing cloud or platform investment, there's a real efficiency case for consolidating rather than adding a 5th vendor relationship. The process check here is honest math, total cost and integration overhead of a unified stack against best-of-breed performance on your top 2 or 3 use cases. Don't assume the bundle wins without running the comparison. The common thread across all 5: pilot before you scale, define the outcome before you start, and choose partners with proof in your specific industry, not just proof in general. What this looks like by function. Choosing the path is only half the work. What it demands looks different depending on which seat you sit in. For marketing and sales, pilot workflow use cases now— content generation, lead qualification— while treating platform choice as reversible for another 12 to 18 months. Operations and supply chain, you have the most to gain from agentic deployment partnerships. Be the first to demand industry-specific references before signing anything multi-year. Finance and IT, build vendor evaluation criteria that weight deployment support and services investment as heavily as model benchmarks. The deployment gap, not raw model capability, is turning out to be the real bottleneck. For investors, the diligence question is shifting. Platform choice stopped being the question. Proof is one live workflow, one measurable outcome, a named reference, and a real pilot are the new signal. A strategy deck with no pilot attached is the exposure. And for vendors, the companies investing in real deployment capacity, engineers who show up and stay long enough to prove out a workflow, are the ones giving buyers an actual answer to which one do I adopt. The shift from seats to outcomes is the strategy that earns that answer. The ones still relying on positioning alone are asking the customer to do work the vendor should be doing. The bet doesn't have to be final to be real. None of this requires waiting for the category to settle down because it isn't going to settle down. on any leader's timeline. What it requires is treating platform selection as a staged, evidence-based process rather than a single irreversible decision. Pilot narrowly, measure honestly, choose partners with proof in your industry, and let the early use cases earn the right to scale. One more thing belongs in that process, and it's easy to skip: setting expectations with the board, the executive team, and employees Before the first pilot launches, not after. A board expecting a magic fix will call an honest 6-month pilot a failure. Employees who hear reimagining workflows without context will assume the worst about their own role, and that anxiety alone can quietly sink adoption before the pilot has a chance to prove anything. Naming the timeline, the scope, and the real stakes up front is what Keeps an early trial from getting derailed by expectations nobody actually set. That's a real strategy. Buying a batch of chat licenses and calling it one isn't. Well, that's the episode. If this was useful, please do one thing. Follow the show on Apple or Spotify and leave a rating. It's the single best way to help other operators find this podcast. I'm Tony Uphoff, back at it next week.