The Next Abstraction on Top of Shannon

In 1948, a Bell Labs engineer named Claude Shannon published a paper called A Mathematical Theory of Communication. It is, by most reasonable measures, one of the most important scientific papers of the 20th century.

Shannon was trying to solve a specific engineering problem: how do you transmit information reliably over a noisy channel? Telegraph wires hummed with interference. Radio crackled with static. Every physical medium corrupted the signal in some way. The conventional wisdom was that more noise meant more errors, and you just had to live with it.

Shannon proved the conventional wisdom wrong. With the right decoder, what we now call error-correcting codes, you could recover a clean signal from an arbitrarily noisy channel, with arbitrarily low error, as long as you transmitted below the channel's capacity. The static didn't matter. The bits could travel through chaos and arrive intact.

This is the foundation of the modern world. Every wifi network, every 5G tower, every TCP packet, every fiber optic cable, every satellite link is downstream of Shannon's idea. Voyager 1, now in interstellar space, still phones home using descendants of his theorems. It's also why the email you sent this morning arrived at all.

But Shannon made one deliberate choice that's worth understanding carefully.

He set meaning aside.

In the opening pages of the paper, he explicitly states that the semantic aspects of communication are irrelevant to the engineering problem. He cared about bits, not what they meant. A message saying "I love you" and a message saying random gibberish were, to Shannon, the same problem: get the bits across.

In 1948, this was a brilliant simplification. The wire was the bottleneck. Solve the wire, and everything downstream would get better. Worrying about meaning before you'd solved transmission was putting the cart before the horse.

So he solved transmission. And for eighty years, an entire industry built itself on top of his work.

The wire is solved

Today, in 2026, the wire is essentially perfect. Bits arrive flawlessly. SMTP delivers your email with effectively 100% fidelity. The cost of transmitting a message has fallen to nearly zero. The reliability of transmission is so high it's invisible. We only notice when something fails, which is almost never.

And yet, somehow, communication has gotten harder.

If you check your inbox right now, you'll find something strange. The bits got through. Every email is technically intact. Headers correct, body uncorrupted, attachments downloadable. By Shannon's standards, the system is working perfectly.

But the message you actually need, the one that requires a decision from you, the deadline you can't miss, the question your client is waiting on, is buried somewhere under 200 newsletters, promotional emails, automated notifications, group threads you were CC'd on out of politeness, calendar invites, and receipts.

Run the numbers on a typical professional's inbox:

• 5% of emails actually need you to do something
• 7% you want to know about
• 88% is noise

That's a signal-to-noise ratio of about 0.14. You are spending hours every week decoding a channel that is 88% static, by hand, with your eyeballs. The data backs this up: the average professional now spends roughly 28% of the workweek on email.

The bits are pristine. The meaning is buried.

The bottleneck moved

This is the part most people miss. We tend to think of communication systems as solved or unsolved. Email "works." Wifi "works." We've moved on to other problems.

But Shannon's framework is more useful than that. Communication systems have layers, and when one layer is solved, the bottleneck moves to the next one up. For most of the 20th century, the bottleneck was physical transmission, getting bits across copper, fiber, and air. By the 2000s, that was done. The bottleneck moved to storage and indexing, and Google solved that. By the 2010s, the bottleneck moved to interfaces, and the smartphone solved that.

Now the bottleneck has moved again, and it's sitting somewhere most communication engineers haven't looked: the human receiver.

We are the noisy channel now.

When 200 emails land in front of a person who has roughly 30 minutes to deal with them, the channel between the sender's intent and the receiver's attention is catastrophically noisy. Not in Shannon's sense, the bits are fine. But in a semantic sense. The meaning is being lost. Decisions are being missed. Deadlines are slipping. Important context is sitting in unread threads. The signal is in there somewhere, but the receiver has neither the time nor the bandwidth to extract it.

And the existing tools can't help, because they're working at the wrong layer.

Why filters were never going to work

Email filters, Gmail rules, SaneBox, server-side sorting, operate at the syntactic layer. They look at sender, subject line, keywords, patterns. They can put newsletters in a folder. They can flag messages from your boss. They can route receipts to a label.

But they cannot read.

A permission slip from your child's school doesn't match a rule. A renewal deadline buried in paragraph four of a long thread doesn't either. A subtle "let's circle back next week if I don't hear from you" that's actually a deadline for you, no filter catches that. Anything requiring an understanding of what the email means, rather than what it superficially looks like, is invisible to filters. I've written about why filters break down at scale, but the deeper reason is that they were always solving the wrong problem.

This is structurally the same gap Shannon faced in 1948. He had no decoder good enough to do what he wanted to do. So he proved mathematically that one could exist, and then engineers spent decades building it. Reed-Solomon codes. Turbo codes. LDPC codes. Each one a more powerful decoder, until eventually we could pull a signal out of essentially any channel.

For the semantic layer, that decoder didn't exist until very recently. Filters were the syntactic equivalent of repetition codes: crude, limited, easily defeated. To do real semantic decoding, you needed a system that could actually read.

That's what large language models are.

LLMs are the first viable semantic decoder. They are to meaning what Reed-Solomon codes were to bits: the first error-correction technology powerful enough to make a new layer of the stack practical. They can read a 40-message thread and tell you the actual decision point. They can find the deadline buried under context. They can distinguish "FYI" from "needs you" in a way that no rule ever could.

This is the unlock. Not "AI in email." Not a feature. A new layer in the communication stack.

Unboxd, in Shannon's terms

This is what Unboxd is, and the cleanest way to describe it is in Shannon's own vocabulary, mapped one layer up:

You don't read the channel anymore. You read the decoded output.

Shannon proved that with a sufficiently good decoder, you can transmit arbitrarily close to channel capacity with arbitrarily low error. The Unboxd bet is the analog at the semantic layer: with a sufficiently good AI decoder, you can extract arbitrarily close to 100% of the actionable signal from a 12%-signal inbox, without the human ever having to see the 88%.

That's a real claim, not a metaphor. It's the same shape of problem, the same shape of solution, just applied one abstraction higher.

Why now

Three things had to be true for this layer to be possible, and only recently have all three become true at once.

First, the wire had to be solved. If bits were still unreliable, you'd be fighting Layer 0, not building Layer 2. They're solved. Email arrives.

Second, the volume had to break the human. If people could still keep up with their inboxes by hand, there'd be no problem to solve. They can't. The average professional spends 28% of their workweek on email and still misses things.

Third, the decoder had to exist. Filters were never sufficient. LLMs finally are.

Each of these was a slow boil. Together, they've created a new layer that didn't exist five years ago and that someone is going to define. Email is the first beachhead, but the same pattern applies to every high-volume semantic channel: Slack, notifications, group chats, document workflows, calendar overload. Anywhere bits arrive perfectly and meaning still gets lost, there's room for a decoder.

The line

I'll close with the way I'd put it in one sentence, because it's the cleanest version of the idea I have.

Shannon taught machines to separate signal from noise on the wire. Unboxd does it for your attention.

That's the bet. One layer up.

Frequently asked questions

What did Claude Shannon's 1948 paper actually solve?

Shannon's A Mathematical Theory of Communication proved that with the right error-correcting decoder, you can transmit information through an arbitrarily noisy channel with arbitrarily low error, as long as you stay below the channel's capacity. He explicitly set the meaning of messages aside and treated communication as a pure bit-transmission problem. Every wifi network, fiber link, TCP packet, and satellite uplink today is downstream of that result.

If email transmission is reliable, why does the inbox feel broken?

The wire is solved. Bits arrive intact. The bottleneck has moved up a layer to meaning. A typical professional inbox is roughly 5% action items, 7% useful information, and 88% noise, so the human receiver becomes the noisy channel. The signal is in there, but extracting it by eye costs hours every week.

Why don't email filters fix the problem?

Filters operate at the syntactic layer: sender, subject, keyword, pattern. They cannot read. A permission slip from your child's school, a renewal deadline buried in paragraph four, a soft ask phrased as "circle back next week" all look like ordinary text to a rule. Anything that requires understanding what an email means, rather than what it superficially looks like, is invisible to filters.

What is Unboxd in Shannon's terms?

The inbox is the channel. Action items, deadlines, and decisions are the signal. Newsletters, promos, CCs, and automated mail are the noise. The AI is the decoder. Your attention is the bandwidth. The daily briefing is the decoded output. You stop reading the channel and start reading the decoded output.