AI as Discovery Engine: What Harrington Is Actually For

You’re not stuck because you can’t write. You’re stuck because you can’t yet see.

Here is a thing that happens to every songwriter at some point.

You have a feeling. It’s specific — not vague, not abstract, not “sad” or “happy” but a particular shade of something that happened on a particular afternoon that you haven’t been able to shake. You know the feeling is real. You know it’s worth a song. And you sit down to write it, and nothing comes out right, and an hour later you have a page of crossed-out lines that each sort of gesture at the thing without landing on it.

That’s not writer’s block in the way people usually mean it — paralysis, self-doubt, a blank page. You have words. You have ideas. What you don’t have is access. The thing you’re trying to say is just past the edge of your current field of vision. It exists. You can feel it from here. But you can’t quite see it.

This is the problem that almost nobody talks about when they talk about AI and creativity. We get distracted by the question of whether AI can replace an artist. The more interesting question — the one that actually matters to working songwriters — is whether AI can help you find what you’re looking for when you don’t yet know what it looks like.

The Map Has an Edge

The economist Thomas Schelling, writing about strategy, described a fundamental limitation of the human mind: “One thing a person cannot do, no matter how rigorous his analysis or heroic his imagination, is to draw up a list of things that would never occur to him.”

He was writing about game theory. But the observation cuts to the heart of every creative stuck moment ever experienced.

Your imagination is not unlimited. It is bounded — not by lack of effort or talent, but by the territory you have already traversed. You can see everything inside your current map. You cannot see what’s beyond the edge. And the song you’re trying to write is often just beyond the edge — not far, not unreachable, but invisible from where you’re standing.

Steven Johnson, writing about innovation, called this the adjacent possible: the set of moves available from any given position, which expands as you explore it. Every step into new territory reveals new territory. The map grows as you travel it. But you have to travel it. Standing still — working harder in the same direction — does not reveal the adjacent possible. It just deepens the rut you’re already in.

Writer’s block, in this light, is often not a writing problem. It’s a navigation problem. You’re not out of words. You’re out of unexplored terrain.

What Artists Have Always Done About This

Long before AI, artists developed methods for breaking out of their own predictability. The history of art is partly the history of tools for outflanking the conscious mind.

Brian Eno is the master practitioner of this. In the 1970s, working in studios with David Bowie, Talking Heads, and others, he encountered a recurring problem: when the pressure was on, musicians defaulted to what they knew. Habit took over. The result was competent but predictable. So Eno, with painter Peter Schmidt, created Oblique Strategies — a deck of over a hundred cards, each bearing a cryptic instruction. “Honor thy error as a hidden intention.” “Use an old idea.” “What would your closest friend do?” “Emphasize the flaws.”

The cards weren’t suggestions for what to play. They were interruptions. Their purpose was to break the grip of the obvious — to force the question “what if I approached this completely differently?” — and see what emerged. Eno described the problem they solved: “The panic of the situation tended to make me quickly forget that there were other ways of working and that there were tangential ways of attacking problems.”

The cards exist because the conscious mind under pressure shrinks. It retreats to the familiar. The tangential move — the one that would actually unlock the song — is invisible from inside that pressure. You need something outside yourself to point sideways.

Igor Stravinsky understood this from a different angle. He argued that constraints — the very things that seem to restrict freedom — are actually what make original work possible. In Poetics of Music he wrote: “The more constraints one imposes, the more one frees one’s self of the chains that shackle the spirit.” He meant that unlimited possibility yields fantasy — vague, unformed, unrepeatable. It’s constraint that forces the artist to discover what’s actually in there. The sonnet’s rigid form doesn’t trap the poet. It narrows the field until something unexpected becomes inevitable.

The Oulipo — a French literary group founded in 1960 and including Georges Perec, Raymond Queneau, and Italo Calvino — built an entire artistic philosophy on this premise. They wrote novels without using the letter e. They created poems that could be recombined into more poems than there are atoms in the universe. They called themselves “rats who construct the labyrinth from which they plan to escape.” The constraint was the tool. The restriction was the engine. The impossible rule revealed what the free mind never would have found.

The Surrealists, a generation earlier, went the opposite direction — not constraint but surrender. Automatic writing, exquisite corpse, chance imagery: methods designed to bypass the editing mind and let whatever was underneath surface. André Breton described automatism not as chaos but as research — “a set of investigative procedures that organize and govern practice but do not determine outcomes.” The goal was the same as Oulipo’s, reached from the other side: find the thing the conscious mind is hiding from itself.

Different methods. Same understanding. The thing you’re looking for is not going to emerge from working harder in the direction you’re already facing. You need to turn.

Hallucination Is the Wrong Word for What Artists Need

Here is where AI enters this conversation in an unexpected way.

One of the main problems with large language models, from an engineering perspective, is that they hallucinate. They generate plausible-sounding content that isn’t true. They invent citations, misremember facts, confabulate details. Engineers work hard to reduce this. It is, in most contexts, a bug to be fixed.

But consider what hallucination actually is, structurally. It is the generation of plausible content not yet verified against the world. The model produces something coherent, something that fits the pattern, something that could be true — but hasn’t been checked.

Now consider what artistic imagination is, structurally. It is the generation of emotionally true content not yet expressed in the world. The artist produces something that fits the feeling — that could render the experience — before they know if it actually does.

These are not the same thing. An LLM has no inner life, no lived experience, no actual stake in what it produces. The equivalence breaks down fast if you push it hard. But the structural parallel — generating the plausible next thing before verification — is real enough to notice.

We tell AI: don’t hallucinate, give me truth and facts. We tell artists: let your imagination run, don’t edit too soon. The discipline we suppress in the model for factual tasks is precisely the discipline we cultivate in the artist. Not because AI imagines the way humans do — it doesn’t — but because the generative, ahead-of-verification quality of hallucination is exactly the gear shift that gets a stuck artist moving.

The songwriter asking Harrington “what’s another way to say this?” is not looking for the right answer. They’re looking for the unexpected adjacency — the thing that would never have occurred to them — that makes them say “wait, that’s not quite it, but that over there…” The AI’s imprecision, its tendency to go slightly sideways, is not a failure. It’s the whole point. You’re not asking it to know. You’re asking it to point.

There’s a second thing AI can do for the working artist that gets even less attention than discovery: it can show you your own habits.

Every artist has blind spots. Not weaknesses — blind spots. The things you can’t see because of how well you see everything else. The word you reach for in every third line. The chord you land on when you’re tired. The metaphor you’ve already used, without realizing, in three different songs. The emotional territory you circle but never enter.

You can’t see these things yourself. Not because you’re not looking, but because you’re you — too close to it, too inside the familiarity of your own voice. It’s not a failure of craft. It’s a structural feature of being the author.

This is what editors, writing teachers, trusted listeners, and workshop partners have always provided: the external eye. The perspective that isn’t inside your head. The observation that “you do this thing every time, and I’m not sure it’s serving you.”

AI can do a version of this. Not with the wisdom or the emotional intelligence of a great teacher or editor — that’s a different thing. But it can read your draft and tell you what you’ve already used. It can analyze the chord progressions in your catalog and show you which move you make most often. It can note that you’ve written about leaving from the outside three times but never from the inside. It can ask you questions about the song that expose what’s missing — the thing you’re circling but not saying.

This is not the AI writing your song. It’s the AI standing outside and reporting what it sees. You still decide what to do with that. But the report can show you the rock you haven’t yet turned over.

Looking for Sparks, Not Answers

The distinction that matters — the one that separates creative AI use from creative AI abdication — is the difference between looking for answers and looking for sparks.

Answer mode: you’re stuck on a line, so you ask the AI for three alternatives and pick the best one. The AI becomes a vending machine. You get a line. You also get a slightly more average version of your song, and a slightly atrophied ability to find the line yourself next time.

Spark mode: you’re stuck on a line, so you ask the AI what a songwriter who hates the approach you’re taking would do instead. Or you ask it to write the verse from the other person’s perspective. Or you ask it what’s missing from the emotional arc. Or you ask it to give you the worst possible version — the cliché so obvious it’s almost funny — and you use it as the thing to push against. You’re not looking for the line. You’re looking for the adjacent possible. The AI doesn’t give you the answer. It changes the shape of the room so you can find it yourself.

Robin Sloan, the novelist, describes writing with AI as working with “a deranged but very well-read parrot on your shoulder.” The parrot doesn’t know what you’re writing. It has no idea what the story is for. But it generates the next plausible thing, and occasionally that thing is so unexpected and so right that it shows you something you couldn’t have seen alone. You keep that. You discard the rest. The writing is yours. The discovery was collaborative.

George Saunders grounds his whole teaching method in something similar — the value of not knowing too well what you’re going to write before you write it. “To know what the story is going to do too well is a buzzkill.” The experienced writer has learned to listen for the story’s own energy, its internal logic, the direction it’s pulling. A tool that introduces something unexpected — that interrupts the too-familiar path — can restore the productive not-knowing that got edited out by habit.

What Harrington Is Actually For

Harrington is built around this understanding.

Harrington’s Spark tool is the clearest expression of it. On an empty line, it doesn’t fill the blank for you; it shows three different things the line could be trying to do. On an existing line, it challenges the assumption underneath what you wrote. One result might be wrong in an interesting way. One might be too obvious, which tells you something. One might land in a place you didn’t know you were headed. None of them is the song. But one of them might turn the right rock over.

Harrington’s other line-level tools protect the same boundary at the appropriate scale. The Suggest tool explores alternatives inside the line-level frame you’ve already chosen. The Refine tool polishes mechanics once the line knows what it wants to be. The Rhyme tool widens the sonic field without pretending that a rhyme is a lyric. At the section level, the Structural Guidance and Chord Progression tools help a section do its job without writing it for you. At the song level, Theme Check and Song Coach help you see what the whole draft is saying. Each of Harrington’s tools is there to make a different kind of possibility visible.

Harrington’s analysis tools aren’t there to judge your work. They’re there to show you what you’re doing that you might not be able to see — the patterns, the repetitions, the gaps in the emotional arc. Not to fix them. To name them. What you do with that is yours.

The master’s perspective isn’t there to write like a master for you. It’s there because the gap between an experienced songwriter and an inexperienced one isn’t primarily about feeling — both feel deeply. It’s about the ability to express feeling precisely. The master has a larger vocabulary for rendering the specific. The master’s perspective can show you that vocabulary, point toward it, make it visible. Then the work of learning to use it is yours.

None of this shortcuts the work. It widens the territory in which the work happens.

That’s what a discovery engine does. Not give you the answer. Not make the choices. Not write the song. Give you access to the terrain just past the edge of your current map, so that when you find the thing you were looking for, it’s unmistakably, irreducibly yours.

The song is always yours. Harrington just helps you find it.

Harrington is a songwriting workbench — built for discovery, not generation. harrington.coach