The “AI slop” problem isn’t a writing problem — it’s a memory problem
If you run a LinkedIn account for yourself or a founder, you already know the real bottleneck isn’t typing. It’s the blank page on a Monday morning, the vague sense that you told that client-war-story three months ago, and the slow creep of every post sounding like every other post in your feed. A new Product Hunt launch, Never Boring AI from maker Maxence VROILANT, is a direct attempt at that specific pain: an AI agent that interviews you first, remembers your anecdotes, and plans your LinkedIn content weeks ahead. It’s a narrow product — LinkedIn only, English and French — but the design choices in it say a lot about where the AI content stack is heading, and where it’s still stuck. Here’s what I think is worth stealing, what I’d test, and where I’d bet it breaks.
The actual problem it solves (and why “generate me a post” was never it)
Most AI writing tools, from ChatGPT to the copy generators bolted onto Buffer and Hootsuite, start from a prompt. You describe a topic, you get a draft. The failure mode is well documented at this point: the drafts are grammatically fine, tonally flat, and interchangeable. The maker’s own framing is blunt about it — “AI tools made posting faster, but the posts all started to sound the same. Generating text wasn’t the problem.”
That’s a real diagnosis, and it’s the same one I’ve landed on after running AI-assisted workflows across Instagram, TikTok, X, and LinkedIn. The tools that help are the ones that solve input, not output. A blank-page problem is not fixed by a faster typewriter. It’s fixed by having a corpus of your own stories, opinions, and half-formed takes sitting somewhere the model can reach.
Never Boring AI’s pitch is that it does exactly that. Per the launch page, the workflow is: you talk to it about your work, it “digs for the real story and keeps your anecdotes in memory,” it plans content several weeks ahead, it writes in your voice and learns from every edit, and nothing publishes without your approval. Scheduling and publishing to LinkedIn happen inside the app, with performance stats afterward.
The “asks questions first” mechanic is the part I’d single out. It’s a small UX decision with a big downstream effect: instead of you prompting the model, the model prompts you. That’s closer to how a good ghostwriter or content strategist actually works. They don’t ask “what do you want to post about?” They ask “what happened last week that annoyed you?” The output is only as differentiated as the input, and most tools never bother to improve the input.
Why this matters more on LinkedIn than on TikTok
Platform mechanics change how much a memory-based agent is worth. On TikTok, the algorithm rewards watch time and rewatches; a hook that lands in the first 1.5 seconds does more work than a clever turn of phrase. On Instagram, saves and shares carry distribution weight that a text-first tool can’t touch. On LinkedIn, the feed still rewards dwell time and comment velocity on text posts, which means the quality of the written idea is closer to the actual lever. That’s why a LinkedIn-only writing agent is a defensible wedge. It’s also why I’d be cautious about anyone trying to port the same model to short-form video without a serious rethink — the input that makes a good LinkedIn post (a specific professional anecdote with a clean lesson) is not the input that makes a good TikTok (a visual moment, a reaction, a pattern interrupt).
How it stacks up against the incumbents
The honest comparison set here is crowded, and I want to be specific rather than hand-wave.
- Taplio and Shield are the closest direct competitors — both are LinkedIn-focused content and analytics tools with AI drafting layered on. They’ve been around longer and have deeper analytics surfaces.
- Buffer, Later, Metricool, and Hootsuite are the scheduling layer. They solve publishing and cross-platform calendars, not voice.
- Jasper and Copy.ai are the general-purpose AI writers. They’re powerful but they don’t remember your life.
- Canva and CapCut own the visual and video side, which this product doesn’t touch at all.
What Never Boring AI is betting on is that memory plus interview is a bigger differentiator than more templates. In my experience testing tools in this category, that bet is directionally right but harder to execute than it sounds. Memory is easy to demo and hard to keep clean — a model that remembers the wrong anecdote, or remembers it slightly wrong, produces drafts that feel uncanny rather than personal. The maker says the product “keeps your anecdotes in memory” but doesn’t say how memory is scoped, edited, or deleted. That’s a gap I’d want closed before I trusted it with six months of client stories.
On pricing: not disclosed on the launch page. On user counts: not disclosed. On funding: not disclosed. I’d rather write “unknown” than guess.
The one detail that actually impressed me
Buried in the comments is the most technically interesting claim in the whole thread. A commenter, Simon Moxon, pushes on the tell that gives AI posts away — the stock phrasing, with “the hard part was never X, it was Y” as his example. The maker’s reply: “I banned all the phrasing that sounds too much like AI from the product’s core code,” then trained the agent on a large corpus of posts to distinguish what works from what doesn’t.
That’s a step past what most tools do. Most just tell you to “write in a human tone” and call it a day. A denylist at the code level is at least a concrete, testable intervention.
Where the denylist logic breaks
And here’s where I’d push back, because Gal Dayan does in the same thread and the exchange is the most useful part of the page. Banning specific phrases is a whack-a-mole problem. Kill “the hard part was never X, it was Y” and the next cliché moves in to fill the vacuum — the short punchy line, the bolded takeaway, the “here’s what I learned” close. Dayan’s framing is sharp: does differentiation just move from “sounds like AI” to “sounds like a slightly-better template”?
The maker’s answer is revealing. He concedes the point about feed-level convergence — “in the LinkedIn feed, we’re bound to see the same patterns, because that’s what works for reach” — and pivots to customization depth, claiming “it’s statistically impossible for two people to have the same type of post” given the settings available. I’d treat that as a maker’s claim, not a verified fact. It may be true at the level of individual posts. It’s almost certainly not true at the level of rhythm, which is what Dayan was actually asking about, and which the maker didn’t fully answer.
My take: the denylist is a good first move and a bad final move. The durable version trains against structure and rhythm — sentence length distribution, paragraph cadence, how the post closes — not against a list of banned strings. Otherwise you’re just moving the template one layer deeper.
What creators and social teams can steal from this, regardless of whether they buy it
Even if you never open the app, three workflow ideas here are worth lifting into your own stack this week.
1. Interview yourself before you brief an AI. Instead of opening ChatGPT and typing “write me a LinkedIn post about leadership,” open a doc and answer three questions: what happened this week that surprised me, who disagreed with me and why, and what did I change my mind about. Then feed that to the model. The quality of your AI output is capped by the quality of your input, and most people skip the input step entirely. This is the single highest-leverage change I’ve made to my own AI-assisted workflow.
2. Keep a story bank, not a content calendar. A calendar tells you when to post. A story bank tells you what to post. In my own tests of similar tools, the ones that work keep a running list of anecdotes, screenshots, and one-liners that you can pull from when the calendar slot arrives. If your tool doesn’t do this, a plain Notion database or a Google Doc does the job. The point is to decouple capture from publishing.
3. Plan in multi-week blocks, edit in single-post sprints. The maker’s claim that the tool “plans your content several weeks ahead” is the part I’d copy even without the tool. Batching planning and separating it from editing is how you avoid the Monday blank page. It’s also how you catch thematic repetition — three posts in a row about the same client, same lesson, same structure — before it ships.
Why TikTok and Instagram creators should care less (for now)
I want to be honest about scope. This is a LinkedIn-first, text-first product. If your primary channels are TikTok and Instagram Reels, the direct utility is low — you’d be using it to draft captions and hooks, which is a fraction of the work. The indirect lesson — memory plus interview beats prompt plus template — applies everywhere. But if you’re hoping this replaces your CapCut workflow or your Metricool cross-posting setup, it doesn’t. It’s a writing agent, not a content OS.
Where I think it falls short (and who it’s not for)
A few things I’d flag before anyone signs up.
LinkedIn-only is a real constraint, not a feature. If you publish across five platforms — which most operators I know do — you’re now maintaining a second workflow just for LinkedIn. That’s a tax. The maker hasn’t announced cross-posting to X, Threads, Facebook, or Pinterest, and until that exists, this is a point solution.
The “external API from LinkedIn” claim needs scrutiny. Asked directly how publishing and stats work, the maker says: “To drive data-driven decisions and get the right statistics, we use an external API from LinkedIn thats much more comprehensive.” That’s vague. LinkedIn’s official API has well-known rate limits and permission scopes, and third-party scraping layers sit in a legally grey zone that has burned plenty of tools before. I’d want to know exactly which API, what the rate limits are, and what happens to my data if LinkedIn tightens access. Not disclosed on the page.
Memory without controls is a liability. For solo creators, remembering your anecdotes is a feature. For agencies running multiple client accounts, it’s a compliance question. Can you isolate memory per client? Export it? Delete it? The page doesn’t say. Until it does, I wouldn’t put client work through it.
The “statistically impossible” claim is marketing, not math. Two users with different settings can still produce structurally identical posts. The customization depth may be real; the statistical framing is not something I’d repeat as fact.
Who it’s not for: teams that need multi-platform scheduling, video-first creators, anyone who needs granular analytics beyond LinkedIn, and anyone who wants to publish without human review at scale. The maker is explicit that “nothing goes out without your approval” — which is the right default, but it also means this is a drafting accelerator, not an autopilot.
What I’d watch / test next
If you want to pressure-test this category — this tool or any of its competitors — here’s what I’d actually do this week.
- Run a blind test on your own audience. Take five posts this tool drafts and five you wrote yourself. Strip the names. Ask three people who know your voice which ones sound like you. That’s the only test that matters, and it’s the one the maker is implicitly asking for in his closing line: “do they sound like you?”
- Audit the memory. Feed it three real client stories, then ask it to write a post. Check whether it used the stories accurately, partially, or invented connective tissue. Accuracy of recall is the whole ballgame.
- Watch the rhythm, not the phrases. After 20 drafts, don’t look for banned clichés. Look for structural sameness — same opening line length, same close, same paragraph count. That’s where the next wave of “AI tells” will live.
- Pressure-test the API story. Ask the maker directly which LinkedIn API, what the rate limits are, and what happens on a permissions change. If the answer is vague, treat analytics as best-effort.
- Decide if a point solution earns a slot. If LinkedIn is 70%+ of your distribution, a dedicated agent can be worth it. If it’s 20%, the workflow lessons above are the real value, and you can implement them with the tools you already pay for.
The interesting thing about this launch isn’t the product itself — it’s the argument underneath it. The next wave of AI content tools won’t win by generating faster. They’ll win by remembering better, interviewing harder, and knowing when not to sound like a template. Whether Never Boring AI is the one that gets there is an open question. But the diagnosis is right, and that’s more than most launches can say.
