Who Is Mike Taylor?
Mike Taylor is the kind of person who makes you rethink what "qualified" means. He started in economics, built a 50-person marketing agency, taught himself to code through a bootcamp in 2020, and by the time O'Reilly needed someone to write the book on prompt engineering, he was one of the few people on the planet with multiple years of hands-on experience. Now he's writing his second O'Reilly book — this one on context engineering with DSPy — while running Ask Rally, a company building AI focus groups with synthetic personas.
His path matters because it contradicts the assumption that the people shaping AI need to come from Stanford CS or Google Research. Mike came from the world of A/B tests and growth marketing. He learned to code so he could talk to developers. He started using GPT-3 in 2020 because he was curious, not because it was his job. Four years of quiet experimentation later, he was the person publishers called.
What makes Mike unusual isn't just the non-traditional path — it's what he does with it. He sits at the intersection of marketing, engineering, and AI research, and he's fluent in all three languages. When he describes DSPy optimization, engineers follow. When he describes it as "a manager checking Asana," marketers follow. That translation ability is the thread connecting everything he's built.
The Archetype: The Sage
The Sage
The Creator
The Reward
Mike's core drive is understanding how things work — and then making that understanding accessible to others. In our conversation, every answer followed the same pattern: framework first, then one perfect story to make it stick. When asked about prompt engineering tips, he didn't share a trick — he laid out a system. When asked about the permanent underclass theory, he didn't share an opinion — he walked through four structural arguments, then compressed the whole thing into seven words.
His secondary archetype is The Creator — underneath the teaching is a compulsive builder who can't stop making things. The marketing agency, Rally, the persona optimization system, a self-cloning experiment. He described it with characteristic self-awareness: "I basically have too many hobbies and interests and it's pretty hard to pin down exactly what I do, but somehow it all works together."
The Sage-Creator combination is what makes Mike's work land with technical audiences. He doesn't just explain — he builds the thing, understands why it works, then writes the explanation. The books aren't theory. They're field reports.
"Almost everything comes down to do you have enough good examples of the task that you're trying to solve. Everything else is incidental to that."
The Hero Match
Daedalus
In Greek mythology, Daedalus was the master craftsman who solved impossible engineering problems through ingenuity rather than brute force. He built the Labyrinth for King Minos, crafted wings from wax and feathers to escape it, and was known for making elegant solutions from whatever materials were at hand.
Mike maps to Daedalus in a way that feels specific rather than generic. He's the craftsman who builds things for others — the agency automated systems for clients, Rally builds synthetic personas for researchers, the O'Reilly books teach engineers. His craft serves other people's needs. Like Daedalus, he's famously practical — not chasing glory, just solving problems with what's available. His entire approach to prompt engineering is about getting maximum results from constrained context windows, limited examples, and cheaper models.
And the deepest parallel: Daedalus's greatest invention was born from constraint. Trapped in his own Labyrinth, he had to invent his way out. Mike's career follows the same pattern — couldn't code, so he taught himself. Didn't have ML credentials, so he built practical expertise through years of brute-force experimentation. The constraint is always the catalyst.
Q (Ben Whishaw version) — James Bond franchise, Skyfall onwards
The modern Q is understated, doesn't look the part, and outperforms everyone in the building without raising his voice. Mike reminded me of this Q throughout our conversation — the technical authority with no flashy credentials, just quiet competence.
When someone at an AI conference told him "hasn't nobody told you prompt engineering is dead?" his response was to write the O'Reilly book on it. That's Q energy — proving capability through output rather than arguing about credentials. The humor matches too: dry, self-deprecating, delivered with a straight face while everyone else takes themselves seriously.
"My most popular prompt is probably just 'fix this' and then a screenshot."
The Story Behind Ask Rally
The story of Ask Rally starts with a confession. At his marketing agency, Mike's team was excellent at A/B testing — running experiments, calculating statistical significance, tracking results. But they rarely did market research. They came from performance marketing, where the assumption was that you could test everything. The problem is that testing is expensive, and a lot of bad ideas could have been killed earlier if someone had just asked the customer first.
When Mike started working with GPT-3 in 2020, every prompt he wrote started with "as this person, you're going to do this task." He was already building synthetic personas before the term existed. One day, the pattern clicked: if AI is already role-playing as customers in his prompts, why not formalize it? Why not build a system that could simulate customer reactions before you build the product?
That experiment became Ask Rally — a platform where you can run AI focus groups with synthetic personas calibrated against real human data. The system creates an LLM judge trained to distinguish real from AI responses, then trains the persona prompt until it fools the judge. Classic adversarial pattern, applied to making synthetic customers more human.
Mike was candid about where Rally stands. "We haven't found product market fit, I'll be perfectly honest. We have about 50 customers right now." He described the market research community's resistance to AI, the challenge of pricing against free ChatGPT, and the narrow overlap of people who both understand AI and value market research. No spin. No pitch deck language. Just an honest read of where things are.
The Founder's Journey ↔ The Company's Journey
Economics degree → 50-person marketing agency (five years) → taught himself to code in 2020 → freelanced as a developer ("nobody knew I didn't have experience") → got early GPT-3 access → quiet years of prompt engineering experimentation → O'Reilly came calling → wrote the book, launched Rally, now writing book two.
Marketing agency side project → prompt-based persona experiments → formal product (Ask Rally) → 300-person calibrated panel, DSPy-powered persona optimization → 50 customers, $20-$100/month plans → searching for product-market fit in a nascent synthetic research market.
The same Sage-Creator archetype drives both arcs. Mike couldn't stop learning new domains — economics, marketing, coding, ML, prompt engineering. Rally couldn't stop adding capability — panels, uploaded interviews, AI-generated personas, evaluator-optimizer loops. In both cases, the builder's curiosity outpaces the market's readiness. The question isn't whether the capability is real. It's whether the market is ready to buy it.
How Mike Leads
Mike's leadership style is framework-first. At his agency, he built accountability systems around Asana — "if it's not in Asana, it didn't happen" — and described debugging people problems the same way he debugs AI pipelines: follow the paper trail, find where it broke, build guardrails so it doesn't happen again.
He applies the same approach to AI. His thesis — that managing AI agents requires managerial skills more than data science skills — comes directly from running a 50-person team. "Did I provide a clear brief? Did I get it to plan first and accept the plan before I let it go ahead? Did I divide the labor up between different workers so that one doesn't get too confused or overloaded?" The analogy isn't forced. It's the way he actually thinks about the problem.
His decision-making is direct without being aggressive. When discussing Rally's challenges, he didn't hedge or redirect. "We haven't found product market fit, I'll be perfectly honest." When assessing competitors, he said they're "really good" — no qualification. This combination of clarity and generosity is unusual in founders — most either minimize competition or minimize their own problems.
Founder Superpowers
Translating between worlds nobody else bridges
Mike sits at the intersection of marketing, engineering, and AI research — and he's fluent in all three languages. At his agency, "I was usually the guy who they would bring in to say, 'hey, Mike, how does this work?' Or 'the developers tell me this, I don't really understand what they're saying.'" That translation ability is why a marketing guy writes engineering textbooks that engineers actually read.
Making the abstract feel inevitable
Mike compresses complex ideas into sentences that change how you think about the problem. "LLMs are a search problem, not a generative problem" — suddenly hallucination isn't mysterious, it's a data coverage issue. "Humans are AGI for minimum wage" — an entire economic argument in seven words. Each reframe makes you feel like you should have already known this. That's the mark of a natural teacher.
Building credibility through demonstration, not credentials
No CS degree. No ML PhD. No Stanford pedigree. "Nobody knew that I didn't have any experience. All they cared about was whether the code worked." Mike applied this approach to everything: years of practical experience before anyone was looking, then the O'Reilly books as proof of mastery. In a field where credential-signaling is the norm, he built authority through output alone.
What It's Like to Work with Mike
Mike is a natural teacher — the kind of person who explains complex systems in a way that makes you feel smarter, not smaller. In an 83-minute unscripted conversation, he never once used jargon to impress. When he described the evaluator-optimizer pattern for persona creation, he immediately connected it to GANs in a way that gave the concept context. When he talked about DSPy, he acknowledged it was "very academic" and "intimidating" before walking through how he actually uses it.
His energy is steady and warm rather than intense. He doesn't dominate a room — he teaches within it. His humor shows up naturally and usually at his own expense. He credits others by name (his co-author James, DSPy creator Omar, even competitors) without being prompted. When asked about Rally's weaknesses, he was more candid than most founders are with their own investors.
The quality that would matter most to a prospective collaborator or team member: Mike is honest about what he doesn't know. "That is unclear to me. I'm not in a good position to understand that" — most founders can't say that sentence. Mike said it about his own market and kept going.
"The imposter syndrome is slowly going, but it's still there. You just keep proving to yourself that you're good enough."
Why This Matters (For You)
If You're Looking for Better Ways to Validate Ideas Before Building
Mike's argument for synthetic research is rooted in a practical problem: testing is expensive, and most teams skip validation entirely because they can't afford focus groups. Rally's approach — calibrate AI personas against real human interviews, then run instant focus groups at $20-$100/month — is one answer. But the broader insight matters even if you never use Rally: AI can eliminate your worst ideas before you waste months building them. "It's not about predicting success. It's about how can I get more lucky? How can I avoid doing obviously wrong things?" That reframe applies to any product team making decisions without enough customer signal.
If You're an Engineer Building AI Applications
Mike's frameworks are immediately actionable. The 10/200 example ladder: under 10, add them to the prompt; over 10, use a prompt optimizer; for production, aim for 200. The search-not-generation mental model: treat LLMs as retrieval systems where your job is making the task more common for the model. The evaluator-optimizer pattern for calibrating outputs. And the single most important habit: look at the logs religiously. "It's actually pretty easy to get lost in the abstraction and not understand what the framework is actually sending." Every one of these is something an engineer can apply tomorrow.
If You're Early in Your Career
Mike's path is the strongest argument against credentialism in AI. Economics degree, marketing agency, self-taught coder, freelance developer, O'Reilly author — none of those steps required permission from a gatekeeping institution. "Nobody knew that I didn't have any experience. All they cared about was whether the code worked." His three career tips are specific: (1) Force yourself to only do tasks with AI, even when it's painful. (2) Run actual Turing tests — send people AI-written vs human-written work and track if they can tell. (3) Have fun with it — make space for play. The senior people are the worst AI adopters because they're already pretty good. You don't have that disadvantage.
If You're Considering Joining Ask Rally
Mike leads with frameworks, credits his team, and is disarmingly honest about where the company stands. "We haven't found product market fit, I'll be perfectly honest" is not something most CEOs say publicly. His decision-making is direct, his humor is dry, and he treats the synthetic research market as a genuine intellectual problem worth solving — not just a business opportunity. The company is early (about 50 customers, $20-$100/month plans), which means high ambiguity but also high agency. If you want a leader who'll be honest about what's working and what isn't, this is the profile.
Go Deeper
The full conversation with Mike Taylor is on its way. Check out other episodes in the meantime.
Read Mike's book
Prompt Engineering for Generative AI — future-proof inputs for reliable AI outputs, co-authored with O'Reilly.