Something’s been bothering me the past few days. I couldn’t put my finger on it at first. It wasn’t anything big or obvious—more like background noise that had suddenly gone quiet. It took two or three days before I realized the absence in my inbox: LinkedIn.

Gone were the frequent job alert emails.

No “someone commented on your post” nudges pulling me back to a conversation I’d started.

No notification that a connection shared something worth reading.

It was an impromptu silence. The kind that makes you do a double-take, check your spam folder, and then sit with this mild but persistent feeling that something changed and nobody told you.

I’d been using LinkedIn with real intention, posting content for THT, keeping up with the dev community, and watching for roles.

It was part of my routine. Not obsessively, but genuinely.

I’d open it and actually find things:

• Interesting projects from various developers
• Industry conversations worth jumping into
• Job postings that felt fresh and relevant

It had a pulse. But that pulse is quieter now.

So I went looking for answers, and what I found was bigger than I expected 🤯

LinkedIn didn’t tweak an algorithm.

It replaced its entire brain, and it did so without anyone’s knowledge.

LinkedIn’s 2026 AI Overhaul: What Actually Changed

Before we get into the specific ways LinkedIn changed, here’s what I found when I started digging.

In 2026, LinkedIn scrapped thousands of individual task-specific ranking models and replaced them with a unified 150-billion-parameter AI system called 360Brew.

Each of those task-specific ranking models handled a specific piece of the platform.

They paired 360Brew with a sequential ranking engine called the Generative Recommender (GR).

This wasn’t a tweak or a refresh.

The entire foundation of how LinkedIn decides what you see, who sees your content, and what jobs to surface was rebuilt from scratch.

The impact was immediate and measurable:

• Organic reach for company pages dropped 60–66%
• Individual visibility fell 47% year-over-year (2025 vs. 2026)
• Follower growth declined 42%
• Overall engagement dropped 39%

Those aren’t rounding errors. That’s a platform-wide shift, and people felt it.

From founders to salespeople and developers, everyone was reporting the same thing: content that used to get traction was flatlining. Posts going quiet. Feeds getting stale. The whole thing feeling less alive.

The question showing up everywhere was the same: Is LinkedIn broken?

Turns out the answer isn’t that LinkedIn is broken. It’s that LinkedIn is different, and it changed the rules without telling anyone.

Let’s go over exactly what happened.

1. Why Your LinkedIn Job Alerts Feel Stuck on Repeat

I had a few saved job searches, specific enough that I expected some variety in what came through.

What I started getting instead was the same handful of roles, from the same cluster of companies, recycled across multiple alerts.

Before, there was range. Different titles, different companies, occasionally something unexpected that made me think “hmm, I hadn’t considered that angle.”

That variety is gone. Here’s the why behind it 👇

MixLM: When Precision Becomes the Problem

LinkedIn replaced its job matching logic with something called MixLM—a semantic matching engine with a 450x compression factor.

What that means in plain English: instead of filtering jobs based on keywords and criteria the way the old system did, MixLM converts your entire profile into a mathematical coordinate (an “embedding”) and then finds job descriptions whose embeddings land close to yours in that same mathematical space.

Note: I talked more about embeddings when generating them out of my personal and resume data for my portfolio. Read more about it right here.

It’s trying to match who you are as a professional, not just what words appear on your resume.

Technically, that’s sophisticated.

The problem is that it lingers when it finds a tight cluster of “perfect” matches. It doesn’t explore the edges.

It doesn’t surface the role that’s slightly outside your established pattern, but might be exactly the lateral move or pivot you needed to see.

The algorithm decided it already knows what you want, so it stops looking 😬

Reddit user djprecio described the new AI-driven job search as “ruining the flow” of finding work, because the precision of the matching came at the cost of discovery.

That’s exactly it.

Sometimes, discovery is all that job searching actually needs. It’s that unexpected posting that reframes what you thought you were looking for.

The Filter Removal Nobody Asked For

If you thought you could manually compensate with filters, that got harder, too.

The platform quietly removed or buried deep filters—things like specific posting dates or remote-only toggles— and replaced them with vague AI-generated categories like “Sustainability” or “Gaming.”

Reddit user AdamCaveRoberts called the new filter options “clunky” and said they stripped out the essential tools needed to surface fresh work.

The ability to filter by posting date alone, just to see new listings, used to be basic. Now it’s buried or gone.

I noticed this firsthand. Clicking on a category like “Remote” shows no visible way to filter further and, often, removes some of the new remote roles that were displaying before selecting the category.

The result is a job feed that’s confident it knows what you want, while quietly offering you less of it.

Tip 😒
If your job alerts feel repetitive, try going directly to the Jobs tab and manually searching with your own keywords rather than relying on emailed alert recommendations. The AI-generated alerts are the most affected by MixLM’s narrow matching behavior. Manual searches give you slightly more control.

2. Why LinkedIn Notifications Basically Disappeared

This one hit differently for me because those notifications meant something beyond vanity metrics.

When I’d post a new THT article or a thought about something happening in the dev world, and someone commented, that notification was a pull back into a conversation.

It made me go back, read what they wrote, and reply. Sometimes it turned into an actual back-and-forth with someone I wouldn’t have connected with otherwise.

The notification wasn’t just a number ticking up. It was what kept the platform feeling alive and reciprocal. It was the reason to come back.

That loop is much harder to trigger now, and here’s the specific reason why:

Likes Don’t Count Anymore

The entire engagement signal hierarchy on LinkedIn has been restructured under 360Brew.

A simple “like” is now nearly worthless. It registers so low on the new scale that it barely influences reach, let alone generates a notification worth sending.

What the algorithm actually values now:

• A thoughtful comment of 15+ words carries 15x more weight than a like
• A save drives 5–10x more reach than a like
• Nested comment threads (i.e., real conversations) trigger a 5.2x amplification effect

What this means practically is that the low-friction engagement that used to generate notifications—a quick like, fire emoji, or one-word “great post!” comment—no longer meaningfully triggers the system.

Former LinkedIn insider Dani Markovits put it plainly: low-friction actions barely register anymore because they don’t prove someone actually read the post.

The platform now demands proof of depth, not proof of presence.

360Brew Is Filtering Content Before It Reaches You

There’s a second layer to this, too. 360Brew isn’t just filtering engagement signals; it’s filtering content before it ever reaches people.

The system actively downranks what it identifies as “AI slop” (generic, templated content) and has made engagement pods (i.e., coordinated groups that like each other’s posts to game reach) essentially useless by detecting what it calls “Coordinated Activity Rings.”

So the notification drought isn’t because people stopped engaging, but because most engagement now happens at a depth the old system never required.

The quick like that used to pull you back into a conversation? The algorithm no longer considers it meaningful enough to act on.

Tip 👀
Are posting on LinkedIn and want your content to actually reach people? Then design posts for saves and substantive replies, not likes. End with a real question. Share something specific enough that someone has a genuine reason to write more than five words back.

3. Why Your LinkedIn Feed Feels Like a Time Loop

This is the one that frustrated me most, because the feed used to be one of LinkedIn’s most underrated features for developers.

Before, opening LinkedIn felt like a browse. You’d catch something from a random dev who just shipped a project, or a discussion thread you hadn’t seen, or an article that reframed how you were thinking about something.

The feed had texture and surfaced things you didn’t know you wanted to find.

That’s actually hard to do well, and for a while, LinkedIn did it reasonably well.

Now I scroll and recognize everything. Same topics, same faces or logos, same conversations from last week. I close the app faster than I opened it.

Two interconnected things are causing this, and they’re worth understanding separately before seeing how they compound.

FishDB and the 30-Day Hard Window

LinkedIn’s retrieval engine, called FishDB, enforces a strict 30-day window on connection-based content.

Anything older than that simply cannot surface in your feed. It’s not deprioritized; it’s excluded entirely.

Within that window, the algorithm has also shifted from recency to relevance.

A high-quality post from three weeks ago that aligns with your interaction history will now outrank something mediocre posted this morning.

On paper, that sounds like an improvement. In practice, it means the content pool the algorithm is pulling from is both narrow in time and narrow in scope.

The Echo Chamber Effect

The Generative Recommender treats your last 1,000 interactions as a chronological sequence and uses “causal attention” to weight them. This means that your most recent behavior carries the heaviest influence on what you see next.

Engage with a few posts about CSS transitions last week? The algorithm reads that as a strong signal, doubles down on it, and starts crowding out everything else.

There’s no natural drift back toward variety unless you deliberately interact with different content.

Reddit user hoya14 (someone with thousands of followers) reported that their feed went “way quieter overnight,” with content no longer reaching even the people who had explicitly chosen to follow them.

That’s the echo chamber effect playing out at scale. It’s not a glitch but a system working exactly as designed, optimizing for what it thinks is signal and filtering out what it interprets as noise.

The uncomfortable conclusion: the feed isn’t broken. It’s a very accurate mirror of your last few weeks of behavior.

The problem is that most of us aren’t curating our LinkedIn interactions with the precision this system now demands.

We browse casually (when we have time) and engage with a range of things. The algorithm reads that range as an inconsistency and proceeds to narrow down the feed to whatever it thinks is your “real” interest.

Note ⚠️
The 30-day FishDB retrieval window means that if you’ve been less active on LinkedIn recently, the algorithm has very little fresh signal to work with. The less you engage, the narrower and more repetitive your feed becomes. This creates a cycle that’s hard to break without intentionally seeking out and deeply engaging with new content outside your usual topics.

4. Why You’ve Stopped Opening LinkedIn the Way You Used To

All that we’ve talked about so far feeds into something that’s easy to dismiss as personal preference but is actually structural.

LinkedIn, as a platform, has quietly become less rewarding to use casually and more demanding for those who want to stay visible on it.

The Dwell Time Trap

LinkedIn optimized for dwell time. The system now measures, literally, how many seconds you spend on a post.

Spend less than 3 seconds on something, and it gets de-ranked.

Stay longer than 15 seconds, and a reach multiplier kicks in.

The entire ranking system is built around the assumption that meaningful professional value comes from depth. From lingering, reading, and actually engaging.

But that’s not how most of us naturally browse a feed.

We scroll, and we skim. We stop when something catches our eye and move on when it doesn’t.

That natural behavior, which LinkedIn used to accommodate just fine, now actively works against the algorithm’s expectations.

Every quick scroll past something is registered as a signal that the content wasn’t worth your time, affecting what gets shown next.

The One-Niche Problem

The Topic Authority Score compounds this. Visibility is now tied to a score built over 60+ days of consistent, focused posting in a single niche.

One niche.

For someone using LinkedIn to post about dev projects, write about the craft, share blog content, and keep up with what’s happening across the industry, that’s not one niche.

That’s a normal, multidimensional professional presence—a variety that the algorithm now penalizes with reduced reach.

I felt this specifically when posting for THT, sharing dev thoughts, and engaging with the community. None of that fits neatly into a single content vertical.

The platform is now structurally less friendly to that kind of authentic, varied engagement.

Users like pallen123, who has 34,000 followers and reported posts reaching only 1,000 people, landed on the honest conclusion that the effort no longer justified the return.

Is that cynicism? No. That’s a rational response to a system that changed its contract mid-game, without warning, and without explanation.

Is This AI Implementation Actually a Success?

Now, let’s take a step back because this isn’t just a LinkedIn story. It’s a real-world, large-scale AI implementation worth examining honestly.

The Case LinkedIn Would Make

From LinkedIn’s perspective, the 360Brew overhaul probably reads as an internal win.

Replacing thousands of fragmented models with one unified system reduces infrastructure complexity and cuts operational costs significantly.

The Generative Recommender and MixLM were designed to deliver precision to stop serving you irrelevant content and instead match you to exactly what the data says you want.

Less noise, more signal. Smarter matching. A leaner, more “intelligent” platform.

And honestly? None of that sounds bad.

Filtering out AI slop and engagement pod manipulation is good.

Rewarding depth over empty likes has real logic to it.

Semantically matching jobs to your actual professional identity, rather than keyword overlap, is a technically sound direction.

The Experience Tells a Different Story

But here’s the part that actually matters: the experience it created for real users is, by most accounts, worse.

Reach collapsed. Engagement dried up. Job discovery narrowed.

Feeds became echo chambers.

People started opening the app less, or not at all.

You can build a technically impressive AI system and still build the wrong thing.

LinkedIn optimized hard for signals it could measure, things like dwell time, comment length, and topic consistency. Yet, in doing so, underestimated something much harder to quantify: the power of serendipity.

Think of the unexpected post.

The job you didn’t know you were looking for.

Or, the conversation that started from a notification you almost ignored.

That’s not noise. That’s actually how professional growth and discovery happen for a lot of people.

AI Is Good for Some Things, Not All Things

AI is getting applied to everything with the implicit assumption that more intelligence always means better outcomes.

Well, it doesn’t 💁‍♀️

And this is an important observation to understand, especially right now.

There’s a certain kind of experience that benefits enormously from AI, like pattern recognition, semantic search, fraud detection, and content filtering at scale.

LinkedIn’s new architecture does some of those things well.

The problem is that it applied that same optimization logic to inspiration, to discovery, to the organic messiness of how people actually build a professional identity.

That’s where it breaks down.

My honest take is that having my feed not entirely tailored to me is one of the few ways I actually get inspired.

Seeing something outside my usual niche—a role I hadn’t considered, a developer doing something I didn’t know was possible, a conversation happening in a corner of the industry I don’t usually sit in—that’s what pushes me to build differently, post differently, and think differently.

When an algorithm decides it already knows who I am and what I need, it stops helping me grow and starts just reflecting me to myself.

That reflection gets narrower every week.

Your LinkedIn presence should be shaped by your curiosity and your choices, not locked in by what an algorithm has decided your “professional embedding” says about you.

Especially as a developer, where your interests, your stack, your direction, and your goals are genuinely in flux most of the time.

We’re not static profiles. We’re moving targets.

A system that anchors on who you were last month isn’t serving who you’re trying to become.

A Big Bet That May Have Missed the Point

At the scale LinkedIn operates over a billion users as a platform that millions rely on for career moves, professional visibility, and industry awareness, this implementation has had a real and measurable impact.

Not catastrophic, but meaningful.

It raises a question worth asking more often as AI gets layered into more products and platforms: Just because you can optimize for a signal, should you?

LinkedIn could measure dwell time—so it did.

It could build a Topic Authority Score, and it did.

It could collapse thousands of models into one. It did that too.

Each of those decisions made some sense individually.

Together, they created a platform that’s technically more sophisticated and experientially less useful for the people it was supposedly built to serve.

That’s worth acknowledging because it’s one of the clearest real-world examples we have right now of what happens when AI optimization runs ahead of user understanding.

For us as developers building things and deciding where AI actually fits, that’s worth remembering.

Related: You’re Not Just Writing Code, You’re Architecting an Experience

What This Actually Means, Especially for Developers

LinkedIn used to fill a specific and valuable role for developers. Not necessarily as our primary community space—that’s always been GitHub, Discord, Twitter/X, dev Slack communities—but something distinct.

The place where you built a professional presence that spoke to non-dev stakeholders. To hiring managers, cross-functional peers, and recruiters.

The platform where your craft had visibility beyond the dev bubble.

That role still technically exists, but the platform facilitating it has changed what it’s optimizing for in a fundamental way.

LinkedIn is no longer a social stream that rewards showing up and engaging organically.

It’s repositioned itself as a Professional Intelligence Tool that expects depth, niche focus, months of consistent behavior, and interactions that demonstrate real attention rather than passive presence.

That’s a very different contract than the one most of us signed up for.

It kinda seems like we now need teams (of humans or agents, your pick) to simply make us seen the way that the algorithm wants to see us.

The Logic Is Sound, The Silence Isn’t

What frustrates me about all of this is that the logic isn’t wrong.

Surface-level engagement doesn’t prove professional value. Depth does.

A post worth saving means more than a post worth liking.

A real conversation thread means more than a fire emoji.

You can see the reasoning.

What’s harder to accept is the silence around it. No announcement or “here’s what we changed and why.”

No acknowledgment that the platform millions of professionals were actively using to find work, build presence, and stay connected with their industry was being rebuilt from the ground up.

It just shifted and left a lot of people standing in a quieter room, wondering what they did wrong.

You didn’t do anything wrong. The room changed.

It’s a Wrap

If your LinkedIn has felt broken lately and you’ve noticed stale job alerts, a dead notification bell, and a feed that serves you the same five topics on rotation, you now know exactly why.

LinkedIn didn’t tweak its recommendation system. It replaced the entire thing with a 150-billion-parameter AI architecture that redefined engagement, relevance, and reach from the ground up.

The casual, exploratory way most of us used the platform by browsing for inspiration, checking job alerts for variety, or getting pulled back by a notification into a conversation was built on a system that no longer exists.

What exists now rewards depth over breadth, niche consistency over authentic variety, and dwell time over casual discovery.

Whether that’s a better platform depends entirely on how you’re using it.

For career positioning and authority-building in one area, the new system has a logic to it.

For the rest of us who just wanted a professional platform that felt alive and helped us discover what we didn’t yet know we needed, the loss is real. You’re not alone in feeling it.

The bigger takeaway is that LinkedIn’s 2026 overhaul is a live case study in what happens when AI optimization and user experience diverge.

It’s a reminder that intelligence in a system doesn’t automatically mean the system serves you better.

As developers who are building things, using AI tools, and deciding where and how to apply them, that’s a lesson worth noting.

You’re not imagining it. The platform changed, and it didn’t tell you.

Has your LinkedIn experience shifted in ways I didn’t cover here? I’d love to hear what you’ve noticed in the comments.

I’ll see ya, bye.