Over the last year, I’ve spent a lot of time thinking about where AI actually fits inside real work — not in theory, but inside the systems teams use every day.
At StitchDX, we work inside CRMs, intranets, knowledge bases, and customer-facing workflows. We see organizations eager to adopt AI, but often unsure how to integrate it without creating inconsistency, risk, or more noise than value. The tools are available. The clarity usually isn’t.
That tension pushed me to go deeper myself. Beyond reacting to headlines or testing isolated tools, I started experimenting directly with the kinds of workflows we help customers navigate — content systems, research and curation, customer-facing automation, and internal knowledge sharing.
Some teams have done the work to deeply understand how AI fits into business environments, and their thinking has shaped a lot of my own learning. But many organizations are still guessing. I was too. Working through that uncertainty inside real systems — with real constraints — has been far more instructive than any list of “best practices.”
Problem Solving
One of the most immediate ways AI has improved my day-to-day work is by reducing friction when something breaks.
Whether it’s a workflow producing unexpected outputs, a prompt behaving inconsistently, or a technical issue I don’t immediately know how to diagnose, AI has become a useful second set of eyes. It helps surface where logic breaks down, highlights inconsistencies in inputs, and shortens the time between noticing a problem and moving forward.
On a broader level, this kind of faster problem-solving matters because so much of the work we do at StitchDX happens inside shared systems — CRMs, intranets, and customer-facing tools that multiple teams rely on every day. When issues are identified and resolved quickly, it becomes easier to keep workflows, content, and internal tools operating consistently across teams.
It’s not about AI fixing things for me. It’s about seeing problems more clearly — and preventing small breakdowns from compounding inside larger systems.
Brainstorming Content Ideas
Starting is often the hardest part of writing — a social post, an email, a blog, or something longer. Staring at a blank document with unlimited possibilities can feel paralyzing.
In a business context, though, the challenge isn’t just getting started — it’s staying aligned. When different people write from different starting points, tone, terminology, and emphasis can drift across customer-facing content.
This is where more intentionally built content GPTs have made a meaningful difference. Instead of acting as one-off writing assistants, shared GPTs can be grounded in the same customer context, language, and reference points. That means ideas are generated from a consistent foundation, even when different people are using them.
In practice, this helps teams move faster and stay aligned. AI isn’t deciding what to say, but it helps ensure that content across emails, blogs, and other touchpoints reflects a shared understanding of customers, problems, and priorities — something that’s difficult to maintain at scale without shared systems.
Multi-Step Workflows Using AI Partner Appy
Over the past year, I also went from knowing very little about AI workflows to understanding how orchestrated, multi-step systems behave in practice.
Working with Appy.AI, we built agents designed to coordinate multi-step workflows.
Each workflow assigns specific tasks to sub-agents — such as research, validation, formatting, storage, and delivery — all managed by an orchestration layer.
The orchestration layer manages handoffs between sub-agents, each with its own clearly defined prompt and responsibility. Together, these agents support research, structured content creation, and consistent output formats while following rules and constraints defined up front.
That’s where things got interesting.
Finding What Worked (and What Didn’t)
As workflows grew more complex, it became clear how quickly AI systems unravel without clear guardrails in place.
We had separate prompts defining behavior, search logic, formatting requirements, and validation rules.
In some cases, we asked the system to prioritize too many things at once — and the cracks showed quickly.
Content would fall within the correct date range but miss the intended topic entirely.
Tightening one filter often created unexpected issues elsewhere.
I initially provided partial outlines rather than strict formats, which led to inconsistent outputs. One newsletter included three bullets per article. The next added publication dates and sources we never asked for. Individually, each output was fine. Collectively, the system lacked consistency.
On the other end of the spectrum, I made the rules too rigid. When I manually provided content we already knew was approved, the system still ran it through every validation step.
In some cases, it rejected the content altogether.
We eventually found a better balance by loosening constraints where judgment mattered more than automation, and by giving humans clearer control points.
The takeaway wasn’t that AI failed — it was that AI amplified whatever ambiguity already existed.
\What This Looks Like Inside Real Organizations
These patterns aren’t unique to my own workflows. We see them repeatedly across organizations experimenting with AI.
Teams want to launch AI chatbots, but no one owns the source content — so answers drift or conflict. Different employees use AI tools with different tones, assumptions, and reference points, creating inconsistency across customer-facing materials. AI outputs get labeled “unreliable,” when the real issue is inconsistent or incomplete inputs.
Employees are already using generative AI tools, with or without formal approval. Ignoring that reality doesn’t reduce risk — it increases it. This is where concerns around data safety, compliance, and shadow AI start to surface.
How StitchDX Thinks About AI in Practice
This is why our approach to AI at StitchDX starts well before automation.
We focus on centralizing trusted content, understanding where sensitive or overshared data lives, and establishing clear ownership, guardrails, and policies around how AI is used — and where it shouldn’t be.
Whether that work happens inside an intranet, a CRM, or an internal knowledge platform, the goal is the same: help organizations use AI confidently and responsibly, without sacrificing consistency, security, or trust.
Looking Ahead
Going into 2026, I have a much clearer sense of how AI fits into real workflows.
I spend more time thinking about structure before automation, clarity before scale, and inputs before outputs. AI has helped reduce friction, create more consistent touchpoints, and open up new ways for organizations to share knowledge and expertise — but only when expectations are clear and ownership is defined.
I’m still learning. But I’m learning with better guardrails, stronger systems, and a deeper appreciation for where AI helps — and where it needs boundaries. That’s the mindset we bring to our work at StitchDX — helping teams use AI with clarity, structure, and confidence. If you’re exploring similar challenges, let’s talk.