Build on the Stack You Have: The AI Playbook That Actually Works

Build on the Stack You Have: The AI Playbook That Actually Works

At SaaStr AI 2026, three people with very different vantage points sat down separately and gave almost the same talk.

Sharif Mansour oversees AI across Atlassian's 20-plus apps and 450 product managers. Eleanor Dorfman runs commercial and industries sales at Anthropic. Rory O'Driscoll has spent years investing in software at Scale Venture Partners. Different roles, different companies, different incentives.

Same playbook.

That kind of convergence is worth paying attention to, especially if you run a small or mid-sized operation and you're trying to figure out where AI actually fits.

What the Playbook Actually Says

The short version: stop waiting for the perfect AI setup. Start with the tools and data you already have. Measure what changes. Expand from there.

This sounds obvious. It is not how most teams operate. Most teams are either waiting for some future AI product that will do everything, or they're running pilots that never connect to real work. The SaaStr panel pushed back on both.

The core argument, as reported by SaaStr, is that AI value compounds when it's embedded in existing workflows, not bolted on as a separate tool. Atlassian didn't rebuild Jira from scratch around AI. They threaded AI capabilities into the product surface their users already lived in. Anthropic's commercial team found that customers who got results fastest were the ones who didn't over-architect before deploying.

O'Driscoll's investor framing added another layer: the businesses that will win with AI are not the ones with the most sophisticated models. They're the ones with the cleanest data and the tightest feedback loops between what the AI does and what a human reviews.

Why This Applies to Operators, Not Just Enterprise Teams

It's easy to read SaaStr coverage and assume it only matters to Series B startups or enterprise product teams. That's wrong.

The same logic applies if you run a clinic, an agency, a retail operation, or any business where you're handling repetitive communication, scheduling, data entry, or customer follow-up.

Your stack probably includes a CRM, a messaging tool, maybe a spreadsheet or two doing things spreadsheets shouldn't be doing. That's your starting point. Not a blank slate. Not a full rebuild.

The question isn't "should we adopt AI?" It's "which part of what we already do every day is the right first place to apply it?"

The Clean Code Lesson From Shopify

A related data point: The Register reported on how Shopify's engineering team found that AI coding agents performed dramatically better when the codebase was clean, well-documented, and built around explicit contracts between components. The agents didn't need less structure. They needed more.

This maps directly to the SaaStr panel's point about data quality. AI doesn't fix messy inputs. It amplifies them. If your customer records are incomplete, your AI-assisted follow-up will be incomplete. If your internal processes are undocumented, your AI agent won't be able to follow them.

Clean up the inputs first. Then apply the AI.

For most small business operators, that means:

• Standardizing how contacts are created and categorized in your CRM
• Documenting the actual steps your team follows for common tasks (even informally)
• Identifying which communication threads are repetitive enough to template
• Deciding what a good outcome looks like before you automate anything

That last one is underrated. If you can't define success for a human doing the task, you can't measure whether the AI is doing it well.

The Jobs Question Is a Distraction Right Now

The Guardian ran a piece arguing that the AI jobs apocalypse probably isn't coming as fast as predicted, and that the economic disruption may not match the hype. That framing is useful for operators to hold.

The teams getting real results from AI right now are not replacing staff. They're handling more volume with the same headcount, or they're reducing the time staff spend on low-value tasks so they can focus on higher-value ones. That's a different story than replacement.

For a clinic handling 200 appointment reminders a week, automating that communication doesn't eliminate a role. It frees up the person doing it to handle exceptions, complaints, and the calls that actually need a human.

For an agency managing 15 client accounts, AI-assisted reporting doesn't replace the account manager. It means the account manager isn't spending four hours a week pulling numbers from three different tools.

This is the realistic near-term picture. Plan for it, not for a science fiction scenario.

Three Questions to Ask Before You Deploy Anything

Taking the SaaStr panel's logic and applying it practically, here are the three questions worth running through before you add any AI tool to your operation:

1. Does this connect to something we already measure?

If you can't point to a metric that will change when the AI works well, you won't know if it's working. Response time, conversion rate, tickets closed, hours saved per week. Pick something concrete.

2. Who owns the feedback loop?

Someone on your team needs to review AI outputs, at least initially, and flag when they're wrong. Without that, errors compound quietly. This isn't a technology question. It's an accountability question.

3. What's the data quality of the inputs?

Before you automate follow-ups, check whether your contact records are complete. Before you use AI to summarize conversations, check whether those conversations are being logged consistently. Garbage in, garbage out is not a cliche. It's the reason most pilots fail.

Where NuvenarHub Fits This Model

We built NuvenarHub specifically around the logic the SaaStr panel described, though we got there from the operator side, not the enterprise product side.

Most small businesses, clinics, and agencies already live in WhatsApp. Their customers message them there. Their teams respond there. The communication is happening whether there's a system around it or not.

NuvenarHub plugs into that existing behavior. It doesn't ask your team to adopt a new communication channel or train customers to use a new portal. It puts a CRM layer, automation, and AI-assisted responses around something they're already doing.

That's the build-on-the-stack-you-have principle applied to SMB reality. The stack, for most operators, is WhatsApp, a phone, and a spreadsheet. You don't blow that up. You build structure around it.

If you want to see how that works in practice, book a call and we'll walk through your specific setup.

The Honest Limitation

The SaaStr playbook is right about starting with existing workflows. It's less useful on one point: it assumes you have someone internally who can run the feedback loop and make decisions about what to automate next.

At Atlassian, that's 450 product managers. At Anthropic's enterprise customers, there's usually an AI lead or a technical operations person. At a 10-person clinic or a 5-person agency, there often isn't.

That's a real gap. The answer isn't to skip AI adoption. It's to be honest about the bandwidth required and either build it in as a role or bring in outside help for the implementation phase. Trying to run a meaningful AI pilot as a side project on top of everything else usually fails, not because the technology doesn't work, but because no one has time to watch what it's doing.

If you need that external support, our services team can fill it for the deployment and tuning phase, then hand it back once it's running.

What to Do This Week

If the SaaStr consensus resonates and you want to move on it:

• Pick one repetitive task that consumes real hours each week
• Document how it's currently done, step by step
• Identify what data is needed to do it and whether that data is clean
• Find the AI tool (or automation layer) that slots into that specific task
• Set a metric, run it for 30 days, review it with whoever owns it

That's the playbook. It's not flashy. It's what the people running AI at Anthropic, Atlassian, and Scale all said works.

Start narrow. Measure honestly. Expand what's working.