25 Ideas Shaping Our Editorial Roadmap
We spend a lot of time writing things down that never get published. Questions we keep coming back to. Arguments we want to stress-test. Ideas that feel important but not yet fully formed. After a while, those drafts pile up and you realize they're telling you something about what the company actually thinks.
This post is an attempt to make that thinking visible. Below are 25 ideas that are actively shaping our editorial roadmap -- the questions we want to answer in public, the arguments we want to make, and the conversations we want to start. Not all of them are settled positions. Some are things we're still working out. That's kind of the point.
We've grouped them into five themes that reflect where our head is these days.
On Discovery
The word "discovery" gets used loosely in e-commerce. Most of what passes for discovery is really just search with a better UI. These posts are about what genuine discovery actually requires -- and why building for it is so much harder than building a better search bar.
1. The Map Is Not the Catalog Angle: Product catalogs are built around what exists; discovery is built around what a person needs. These are two completely different information architectures, and conflating them is why most recommendation engines feel hollow. Target audience: Ecommerce product managers and engineers Why it matters for Starseek: Positions us as people who have actually thought through the structural problem, not just applied AI to the existing model
2. Why "Customers Also Bought" Is the Laziest Sentence in Retail Angle: Collaborative filtering is a useful tool that became an excuse to avoid understanding shoppers as individuals. There's a specific moment in the early 2000s when the industry decided correlation was the same thing as understanding, and we've been living with the consequences since. Target audience: Anyone frustrated with recommendation systems that feel random Why it matters for Starseek: Establishes what we're reacting against and why our approach is different
3. The Vocabulary Problem: Why Shoppers and Products Don't Speak the Same Language Angle: A shopper says "cozy but not frumpy." A product catalog says "relaxed-fit ribbed-knit crew-neck pullover." Bridging that gap is a linguistics problem as much as a machine learning one, and the industry underestimates how hard it is. Target audience: ML engineers, NLP researchers, ecommerce operators Why it matters for Starseek: Demonstrates technical depth and frames our NLP work as solving a real linguistic challenge
4. What a Great Salesperson Knows That a Search Engine Doesn't Angle: A trained salesperson uses body language, hesitation, and follow-up questions to understand what a customer actually wants. AI assistants can replicate some of this but not all of it. This post maps exactly what gets lost in translation and what can be recovered. Target audience: Retailers, brand builders, anyone thinking about the future of sales Why it matters for Starseek: Grounds our product philosophy in a human analogy that resonates emotionally
5. The Paradox of Infinite Choice: More Products, Less Discovery Angle: As the number of available products has grown exponentially, the quality of product discovery has actually declined for most shoppers. This is counterintuitive and worth explaining carefully. Abundance doesn't help you if you can't navigate it. Target audience: General business audience, ecommerce founders Why it matters for Starseek: Frames our market as a crisis of abundance, not a solved problem
On Trust
Trust is infrastructure. Every good interaction in e-commerce -- a recommendation you act on, a review you believe, a brand you return to -- is built on top of some form of trust that had to be established first. These posts are about what trust actually requires and where the current ecosystem has let it erode.
6. The Review Economy Is Broken. Here's the Ledger. Angle: A specific, sourced accounting of how bad the fake review problem actually is: what percentage of Amazon reviews are estimated to be fraudulent, what the FTC has done about it, which categories are most compromised, and what the downstream effects are on shopper behavior. Target audience: Consumers, brand operators, policy-adjacent readers Why it matters for Starseek: Creates context for Karto without being a product announcement; positions us as the company that has actually looked at the data
7. Why Incentivized Reviews Don't Have to Be Corrupt Angle: The assumption that paying for reviews is inherently dishonest conflates the incentive (compensation for time) with the manipulation (compensation for positive opinions). These are different things, and the distinction matters enormously for how you design a review system. Target audience: Ecommerce operators, consumer advocates, anyone skeptical of Karto's model Why it matters for Starseek: Addresses the obvious objection to our approach head-on and makes the case for why our incentive design is different
8. Trust Scores Are Not Star Ratings: A Framework Angle: Star ratings aggregate opinions. Trust scores assess the credibility of the person giving the opinion. These are orthogonal dimensions that most review systems collapse into one number, which is a significant design error. This post proposes a framework for thinking about both. Target audience: Product designers, review platform operators, data-minded readers Why it matters for Starseek: Positions our trust scoring system as a principled design choice, not a feature
9. The Brand That Bought Its Reviews: A Case Study in Long-Term Cost Angle: A fictionalized but realistic case study of a brand that gamed the review system to accelerate growth, the short-term results, and what happened when the reviews normalized to reality. The point isn't "cheating is wrong" -- it's that the economics don't actually work out. Target audience: Brand founders and operators Why it matters for Starseek: Speaks directly to business self-interest, not just ethics
10. What Shoppers Actually Do When They Don't Trust the Reviews Angle: People haven't stopped researching purchases -- they've migrated to Reddit, TikTok, YouTube, and Discord. This post traces that migration and what it tells us about what shoppers actually want from a review: not polish, but authenticity. Target audience: Brand marketers, ecommerce strategists Why it matters for Starseek: Explains the informal review ecosystem that Karto is designed to compete with on credibility
On Personalization
Personalization is one of the most overused words in ecommerce and one of the least well understood. These posts try to make the concept precise: what it requires technically, where it fails ethically, and what it could look like if it were done well.
11. The Difference Between Personalization and Surveillance Angle: A lot of what gets called personalization is actually just tracking. The distinction matters: genuine personalization requires a model of who you are and what you want; surveillance just records what you did and sells it to whoever bids. Shoppers can feel the difference even when they can't articulate it. Target audience: General ecommerce audience, privacy-conscious consumers, brand marketers Why it matters for Starseek: Articulates our approach to user data as a design philosophy, not just a compliance posture
12. Why Your Recommendations Get Worse After You Buy Something Angle: Most recommendation systems weight recent behavior heavily. So the moment you buy a crib, you get six months of baby product recommendations -- even if you only needed one crib. This is a technical flaw with a specific name (filter bubbles around purchase events) and it points to a deeper problem in how temporal context is handled. Target audience: Ecommerce product managers, curious consumers Why it matters for Starseek: Demonstrates the specific failure modes we're designing around
13. Learning What Someone Doesn't Want Is as Important as Learning What They Do Angle: Preference models are almost entirely built on positive signals -- clicks, purchases, saves. But knowing what someone dislikes is equally informative and far harder to collect. This post argues that building for negative preference signals is the underrated frontier in personalization. Target audience: ML engineers, product designers Why it matters for Starseek: Highlights a specific technical direction in our personalization work
14. The Cold Start Problem Is a Product Design Problem, Not Just an ML Problem Angle: Every personalization system struggles with new users who have no history. The standard response is to collect data quickly. The better response is to design onboarding that makes the cold start itself useful and even enjoyable -- so users are willing to share the preferences that make the system smarter. Target audience: Product designers, founders building personalized products Why it matters for Starseek: Positions our onboarding philosophy as a competitive advantage
15. What "Taste" Actually Means and Why It's Hard to Model Angle: Taste is context-dependent, contradictory, and changes over time. A person might prefer minimalist design in their home and maximalist expression in their clothing. Any personalization system that treats "taste" as a stable vector is going to produce results that feel superficially right and fundamentally off. This post unpacks the challenge. Target audience: AI researchers, thoughtful product builders, aesthetics-adjacent readers Why it matters for Starseek: Shows the intellectual seriousness we're bringing to a genuinely hard problem
On Building
These are posts about the process of building Starseek itself: the decisions we've made, the ones we've reversed, and the things we've learned that don't fit neatly into a product announcement. This is where we think out loud about what it means to build an AI company in public.
16. Why We're Building Across Multiple Platforms Instead of One Angle: The conventional wisdom for early-stage startups is to pick one channel, do it well, and expand later. We're indexing across 11 platforms on day one. This post explains the specific logic behind that decision and the real trade-offs we accepted. Target audience: Startup founders, ecommerce operators Why it matters for Starseek: Makes our strategic choices legible and positions them as principled, not scattered
17. The Feature We Killed Because Users Loved It Angle: An honest account of a feature in Spark that early users rated highly but that was actually making the product worse over time -- and what it took to decide to remove it. The lesson is about the difference between what users say they want and what actually improves their outcomes. Target audience: Product managers, founders Why it matters for Starseek: Demonstrates the kind of rigorous product thinking that separates companies with a real philosophy from ones that just follow feedback
18. What Building an AI Product Has Taught Us About Our Own Assumptions Angle: A founder-perspective post about the moments in building Spark where the AI surfaced assumptions we didn't know we had -- about how people shop, about what language means, about what "good" looks like in a recommendation. Honest and specific. Target audience: Founders, AI practitioners, product builders Why it matters for Starseek: Builds founder credibility and intellectual honesty at the same time
19. The Data We Don't Collect (And Why) Angle: Most companies describe data collection in terms of what they do gather. Describing what you explicitly choose not to collect, and why, is a different kind of statement. This post enumerates the categories of user data we've decided are not worth having and explains the reasoning. Target audience: Privacy advocates, conscientious consumers, potential enterprise customers Why it matters for Starseek: Builds trust through specificity; this is harder to fake than a general privacy policy
20. On Not Launching Until It's Worth Launching Angle: There's enormous social pressure in startupland to ship fast, get users, and iterate in public. This post argues for a different posture -- one where the bar for "ready" is higher, because in consumer AI, a bad first impression is particularly hard to recover from. Target audience: Founders, investors, anyone building consumer-facing AI Why it matters for Starseek: Explains our deliberately paced go-to-market and signals that we have standards
On the Industry
These posts step back from Starseek specifically and engage with bigger questions about where AI, commerce, and consumer technology are headed. We have opinions. These are the ones we think are worth saying out loud.
21. The AI Shopping Assistant Graveyard: Why So Many Have Failed Angle: There have been at least a dozen well-funded attempts at AI-powered shopping assistants in the last decade. Almost all of them are dead or irrelevant. This post does a forensic analysis of the most instructive failures and extracts patterns: what they got wrong about the problem, the product, or the market. Target audience: Investors, ecommerce professionals, AI practitioners Why it matters for Starseek: Shows that we understand the failure modes in our own space and have thought about how to avoid them
22. Ecommerce Is Not a Solved Problem Angle: Amazon's dominance has convinced a lot of people that the hard problems in online retail have been figured out. This post argues the opposite: that Amazon solved logistics and scale, but essentially nothing about the shopping experience itself has fundamentally improved in 20 years. The surface got polished; the underlying problems stayed. Target audience: Founders, investors, ecommerce operators Why it matters for Starseek: Frames our market as wide open rather than crowded
23. When AI Gets the Recommendation Right But the Explanation Wrong Angle: One of the most common failure modes in AI-driven commerce is a system that surfaces the correct product but provides reasoning that feels hollow or irrelevant. Users reject recommendations that they can't trust, even when the recommendation itself is objectively good. The explanation is part of the product. Target audience: AI product builders, UX researchers Why it matters for Starseek: Articulates why we invest in explainability as a first-order product concern
24. The Aggregator Trap: What Happens When You Build on Someone Else's Catalog Angle: Any company that aggregates products from third-party platforms is at the mercy of those platforms' API policies, data sharing agreements, and competitive interests. This post thinks through the long-term structural risks of aggregation-dependent business models and how we think about reducing that dependency over time. Target audience: Investors, technical founders, ecommerce strategists Why it matters for Starseek: Signals strategic awareness of our own vulnerabilities and how we're thinking about them
25. The Shopping Experience That Doesn't Exist Yet Angle: A speculative but grounded post about what online shopping might look like in ten years if the core infrastructure -- discovery, trust, personalization -- actually gets rebuilt well. Not a prediction, but a design brief for an experience that doesn't exist yet and that we think is worth building toward. Target audience: General business and technology audience Why it matters for Starseek: Establishes the long-term vision in a way that's inspiring without being vague; this is the kind of post that people share because it articulates something they've felt but couldn't say
That's the agenda. Some of these will become full posts. Some will turn into smaller pieces. A few will probably get combined or thrown out as our thinking evolves. But this is where our head is right now -- the questions we keep coming back to, the arguments we want to make, and the conversations we want to be part of.
If something on this list resonates with you or you have a perspective we haven't considered, we'd genuinely like to hear it. Our contact page is the best place to reach us.