October 14, 2026
AI has dramatically accelerated our ability to write code. But choosing the right products to build with AI still relies on human judgment, experience and intuition.
As the speed of engineering increases, the demand for high-quality product decision-making is rising with it. In 2025, Andrew Ng reported that teams in Y Combinator — a large Silicon Valley startup accelerator — proposed staffing two product managers for every engineer to keep up with AI-accelerated engineering output. A respondent to the 2026 State of AI In Product survey summed it up: “Delivery of designs and code got very fast. Delivery of good decisions became the new bottleneck.”
When you can build almost anything quickly, the quality of decision-making matters more than ever. Any poor decisions around which problem to solve or which risks to take are amplified and can cause more damage, more quickly. The danger is that pressure to keep pace with engineering leads product teams to sacrifice quality for speed.
The best product decisions still need to be made by cross-disciplinary teams of humans. These teams bring skills that AI can’t replicate: empathy, instinct and unconscious domain knowledge that combine to deliver incredible innovation, especially when collaborating with others.
The challenge is to use AI to help product teams make high-quality, aligned decisions at the same pace as engineering teams build.
The first instinct of many product managers (PMs) is to use AI to quickly generate the solution briefs, epics and stories they need. Then they can move as fast as engineering, right?
As a product manager myself, that was my first approach, too. In early 2026, I built a set of Claude Code skills that automated much of my discovery workflow, managing problem selection, solution framing, extracting insights from interviews and writing stories. Each skill would read and write plain markdown files that lived in a shared context layer next to the product code.
I’ve learned the hard way that this approach isn’t sustainable. The automation quickly produced usable artifacts to hand off to developers, but they still needed careful review. The faster they were produced, the more pressure I felt to skip consultation with my colleagues — and the less scrutiny the outputs received.
The problem is that AI-generated documentation puts the onus on the PM to manually review everything. And it often contains complex business logic and mental leaps that are easy for an LLM to produce, but difficult for a human to get their head around.
Studies have shown that an LLM can produce this kind of material about 17x faster than humans can read it (Brysbaert 2019; Artificial Analysis 2026). And hours spent reading endless streams of LLM output quickly induce fatigue, with output passed downstream without proper review, risking serious errors.
AI ends up overwhelming, not augmenting, human judgment.
We’ve become pretty good at capturing human intent and knowledge — like product specs, coding rules and guardrails — in forms that LLMs can reliably understand and work from.
But I think the reverse also needs attention. AI outputs need to be available in forms that humans can readily understand and intuitively reshape, with those human inputs then fed back into the system in machine-readable form.
This represents a move toward a two-way partnership between humans and AI. The goal is for AI to become a true partner that is integrated into those activities that humans do best: collaboration, sense-making, intuition, applying domain knowledge and experience.
This kind of partnership keeps humans at the center of product decisions, while AI takes on more of the heavy-lifting at the edges: setting up sessions, capturing output, enforcing governance and so on.
What does this look like in real life? I adapted my original workflow to support this kind of human-AI partnership.
The key premise is that humans are far better at processing visual information than text; we remember diagrams and can more easily collaborate with others around images. That’s why the best product decisions come out of visual, interactive workshops.
What was missing from my workflow was a way to incorporate traditional visual workshops into the product discovery workflow, so my team could more easily collaborate with me, without sacrificing speed or quality.
When I looked at the AI-generated artifacts for my product, I recognized that they contained the same information I would be working on with my team on a whiteboard. For example, in discovery workshops I’ll frequently build an Opportunity Solution Tree (OST) diagram with my team in Miro to examine the landscape of potential software solutions.
The relevant markdown files already existed in my git repository. So I asked Claude Code to connect to Miro via MCP and trained it to use the files to build an OST diagram.
In a real workshop, my team would collaboratively reshape the OST diagram as the discussion evolved. I wanted to find a way to use AI to easily capture those changes so that the interface works both ways: changes humans made to the visual diagram needed to flow back into the machine-readable files, while changes made by AI to those files needed to appear in the human-friendly diagram.
After several iterations, I built a Claude ‘OST’ skill that could keep the two in sync. Now, I could update the diagram and those changes would flow back into the repo files, and vice versa!
This was the two-way interface I had been looking for. AI could work with the structured, machine-readable files it needed, while my team could work together in a visual environment that suited us. We could move things around, debate ideas and reshape the diagram together, with those decisions automatically captured in markdown for downstream agents to work from.
From there, it was relatively straightforward to generalize this skill to other visual workshop artifacts. I created a similar skill for Story Maps and Assumption Maps — two common workshop tools — where AI could create and edit the diagram, as well as translate any human changes to the diagram back to the underlying repo files.
I discovered that the same loop also works well for prototyping. The LLM turns stories and business context into a design brief, then instructs a prototyping tool (e.g. Claude Design or Magic Patterns) to build it via MCP. The human design team can then tweak and optimize the prototype directly in the tool. The LLM can then read the updated prototype and work out which stories, context files and related visual artifacts need to change.
The result is a workflow that uses AI to truly accelerate the work around human collaboration. I can now prepare workshops in minutes, quickly generate new visual artifacts following the same template, and keep the wider product context up to date.
And this approach has other benefits. It provides a natural place to embed organizational governance into the skills that facilitate our workshops, for example, enforcing approval gates before new stories are added to the backlog. And every product decision generates clean, structured input for other agents downstream (e.g. for design, build or verification).
This product workflow is an example of using AI to augment human ways of working, rather than having humans adapt to LLM-friendly modes.
My team can continue doing what humans do well: collaborating visually, debating ideas, applying experience and making judgments. AI supports those activities by preparing, translating, capturing and carrying the results forward.
If you work in product, you are about to be asked to move faster! But speed alone will not be enough. You still need to put the human strengths of judgment, intuition and experience at the heart of your work, using AI to augment and accelerate the processes around them.
I’ve packaged this whole approach as a Claude Code plugin so you can run it yourself. Point it at your team’s next iteration, or follow the pattern to build a loop for a visual activity of your own, and see how it changes your decision-making. Get the plugin from my Claude Code marketplace here.
Mike Mitchell is a Product & Strategy Principal at Equal Experts, helping organizations adopt modern product practices and integrate AI into their product lifecycle in practical, sustainable ways. His current focus is on helping product teams do more meaningful strategic work with AI, both by accelerating the routine work that crowds it out, and by making AI a more proactive, multi-modal collaborator.
Mike brings 35 years of experience in enterprise software, with a foundation in engineering followed by two decades of product leadership. He is a founder of multiple startups, has advised dozens more, and has led multiple enterprise transformations. Just prior to EE, Mike served as a fractional Chief Product Officer, guiding early- and growth-stage firms to deliver value faster through stronger product strategy and operating models. He holds a degree in Computer Science & Engineering from MIT.
All business models have their pros and cons. But, when you consider the type of problems we help our clients to solve at Equal Experts, it’s worth thinking about the level of experience and the best consultancy approach to solve them.
If you’d like to find out more about working with us – get in touch. We’d love to hear from you.
