On September 27, 2026, a small music startup called Thoughtful Things opened a Kickstarter campaign for a device that does something unusual with artificial intelligence: it lets the machine fail, and then plays the failure. The instrument is called Engram, an Engram sampler and groovebox that processes incoming audio through AI models trained to mangle, warp, and occasionally invent sounds that never existed in the source material. Where most AI music tools promise finished songs, Engram promises raw material — glitches, artifacts, and auditory hallucinations that a performer shapes in real time.
The distinction matters. Music hardware has been enjoying a quiet renaissance for years, and AI music software has been enjoying a loud one. Engram sits at the intersection, and that position raises a question worth taking seriously: when a neural network "hallucinates," is the result noise to be filtered out, or a new class of instrument to be played?
What Is Engram? A New Kind of AI Music Instrument
The global musical instrument market surpassed $6 billion in recent years, according to Music Trades and MI sales reporting, with hardware synthesizers, samplers, and grooveboxes consistently outperforming expectations even as software instruments dominated studio workflows. Reverb.com's annual data has repeatedly shown double-digit year-over-year growth in used hardware sales, with drum machines and samplers among the fastest-moving categories. Physical instruments, it turns out, did not die. They multiplied.
Engram arrives into that market as a sampler and groovebox — two categories with deep histories. A sampler records audio and replays it, typically at altered pitches or speeds. A groovebox sequences patterns, usually with drums and bass parts under a player's fingers. Thoughtful Things combines both and adds an AI layer that transforms whatever audio you feed it. The company's Kickstarter campaign, launched in late September 2026, marks its first hardware product.
Crucially, Engram is not positioned as a song generator. The reported description is explicit: it is not "push-button, get-song" technology like Suno, and it is not aimed at producing top-40-ready tracks. It is an instrument. Instruments require performers. That framing places Engram in the lineage of expressive tools rather than automated content machines — a distinction music technologists have drawn repeatedly when assessing where AI belongs in a creative workflow.
How Engram Turns AI Hallucinations Into Sound
When an AI audio model processes sound, it sometimes produces outputs that don't correspond to anything in the input. Researchers call this hallucination; engineers treat it as an error. Engram treats it as a feature. The sampler runs incoming audio through AI processing that mangles it — stretching, fragmenting, smearing — and in some cases generates entirely new sonic material that the source never contained. A voice note might emerge as a metallic drone. A drum loop might spawn a texture that resembles no drum on Earth.
Read next Laika's Wildwood: Stop-Motion Fantasy at TIFF 2026That behavior echoes a long tradition in experimental music. Oval built entire albums in the 1990s by deliberately scratching CDs and letting the skips become rhythm. Tim Hecker's dense, degraded ambient works rely on processing chains that break sounds apart and reassemble them. Circuit bending — physically shorting toy keyboards and game consoles to produce unpredictable tones — became a recognized practice in the 2000s and remains a fixture of experimental electronic scenes. In every case, the artist's role shifted from generating material to curating and directing it.
Engram formalizes that shift into a playable interface. Instead of patching a glitch through a DAW, a musician captures the hallucination at the moment it occurs and sequences it like any other sample. The machine supplies surprise; the human supplies intent. That division of labor is precisely what separates generative AI products — which compress the creative act into a prompt — from instruments, which extend it across performance.
Engram as a Creative Tool: Sampler Meets Groovebox
Samplers and grooveboxes have shaped popular music for four decades. The Akai MPC series, launched in 1988, became the backbone of hip-hop production. Elektron's Machinedrum and Octatrack established a modern template for hardware sequencing. Roland's SP-404, once a budget afterthought, found a second life as the centerpiece of lo-fi and beat-scene workflows. Each of these tools succeeded not because it automated music-making, but because it made a specific creative act tactile and immediate.
Engram follows that logic. The AI processing is not the product's endpoint — the instrument is. A producer can feed audio into the sampler, let the model distort or invent, then capture the result and build patterns around it. The groovebox sequencing layer means those captured hallucinations can become rhythmic structures, not just isolated effects. This is how samplers have always worked: record, chop, replay, arrange. Engram adds a step where the recording itself becomes unpredictable.
That unpredictability cuts against the grain of most AI music software, which optimizes for coherence and polish. Instrument designers who work on expressive tools consistently argue that the value of an instrument lies in how it responds to a performer's decisions — not in how well it eliminates them. A violin that played itself would not be a violin. Engram, on the reported evidence, applies that principle to neural audio processing.
The Broader Trend: AI Errors as Artistic Raw Material
Glitch music has been mainstream-adjacent for nearly thirty years. Alva Noto, Fennesz, and Autechre all built careers on sounds that originated as technical failures — aliasing, buffer overruns, digital clipping. The aesthetic proposition was simple: what systems discard as error, artists can reclaim as texture. Engram's hallucination engine sits directly in that lineage, even though the underlying technology is new.
The AI angle sharpens the stakes. Neural networks hallucinate because they are statistical engines predicting likely outputs, not retrieving verified facts. In text and image generation, that tendency is a liability. In music, it may be a virtue. A model that produces something unexpected from a familiar input is doing, in crude form, what a jazz musician does when they mishear a phrase and improvise around it.
This is why the framing of Engram as an instrument rather than a generator matters beyond marketing. Generative AI tools have triggered intense debate about authorship, royalties, and the displacement of human creators. Instruments — even AI-powered ones — sidestep much of that debate, because the output depends on a human operator's continuous input. The hallucination is a material. The musician decides what to build with it.
What Musicians and Producers Should Know Before Backing the Kickstarter
Kickstarter hardware campaigns carry risk. Delays are common, and instruments funded before manufacturing scale is proven sometimes ship in revised forms. That said, the crowdfunding model has become a standard route for boutique music hardware: Elektron, Critter & Guitari, and numerous Eurorack makers have used it or its equivalents to bring niche instruments to market. Producers evaluating Engram should weigh the usual factors — build quality promises, delivery timelines, and whether the AI processing is central to their workflow or a novelty they'll use twice.
The more interesting question is fit. If your production style depends on predictable, recallable sample libraries, an instrument that hallucinates may frustrate you. If you work in experimental electronic, industrial, ambient, or beat-driven genres where texture and surprise carry value, the premise aligns with how you already work. The reported positioning — explicitly not a song-generation tool — should be read as a signal about intended use. Engram rewards the player who shows up with ideas and reacts to what the machine gives back.
For producers curious about where AI genuinely fits in a hardware rig, Engram represents one plausible answer: not as a replacement for the musician, but as a collaborator that misbehaves on purpose. The hallucination is the point. What you do with it is the music.
Source: The Verge
