Immediate Constancy: Measuring How A lot of Your Intent an AI Agent Really Executes

Spotify simply shipped “Prompted Playlists” in beta. I constructed a couple of playlists and found that the LLM behind the agent tries to satisfy your request, however fails as a result of it doesn’t know sufficient however received’t admit it. Right here’s what I imply: one in every of my first playlist prompts was “songs in a minor key inside rock”. The playlist was swiftly created. I then added the caveat “and no track ought to have greater than 10 million performs”. The AI agent bubbled up an error explaining that it didn’t have entry to whole play counts. It additionally surprisingly defined that it didn’t have entry to some different issues like musical keys, regardless that it had claimed to make use of that within the playlist’s building. The agent was utilizing its LLM’s information of what key a sure track was in and including songs accordingly to its reminiscence. An in depth inspection of the playlist confirmed a couple of songs that weren’t in a minor key in any respect. The LLM had, after all, hallucinated this data and proudly displayed it as a legitimate match to a playlist’s immediate.

Clearly, a playlist creator is a reasonably low-stakes AI agent functionality. The playlist it made was nice! The difficulty is it solely actually used about 25% of my constraints as validated enter. The remaining 75% of my constraints had been simply guessed by the LLM and the system by no means informed me till I dug in deeper. This isn’t a Spotify downside; it’s an every-agent downside. 

Three Propositions

To reveal this idea of immediate constancy extra broadly, I have to make these three propositions:

1. Any AI agent’s verified information layer has a restricted or finite capability. An agent can solely question the instruments it’s been given, and people instruments expose a hard and fast set of fields with finite decision. You possibly can enumerate each discipline within the schema and measure how a lot every one narrows the search. A recognition rating eliminates some fraction of candidates. A launch date eliminates one other. A style tag eliminates extra. Add up how a lot narrowing all of the fields can do collectively and also you get a tough quantity: the utmost quantity of filtering the agent can show it did. I’ll name that quantity I_max.
2. Consumer intent expressed in pure language is successfully unbounded. An individual can write a immediate of arbitrary specificity. “Create a playlist with songs which are bass-led in minor key, post-punk from Manchester, recorded in studios with analog gear between 1979 and 1983 that influenced the gothic rock motion however by no means charted.” Each clause narrows the search. Each adjective provides precision. There isn’t a ceiling on how particular a consumer’s request may be, as a result of pure language wasn’t designed round database schemas.
3. Following instantly from the primary two: for any AI agent, there exists a degree the place the consumer’s immediate asks for greater than the info layer can confirm. As soon as a immediate calls for extra narrowing than the verified fields can present, the remaining work has to come back from someplace. That someplace is the LLM’s basic information, sample matching, and inference. The agent will nonetheless ship a assured outcome. It simply can’t show all of it. Not as a result of the mannequin is poorly constructed, however as a result of the mathematics doesn’t enable the rest.

This isn’t a high quality downside, however a structural one. A greater mannequin doesn’t elevate the ceiling. Higher fashions do get higher at inferring and filling in the remainder of the consumer’s wants. Nonetheless, solely including extra verified information fields raises this ceiling, and even then, every new discipline gives diminishing returns as a result of fields are correlated (style and vitality aren’t unbiased, launch date and tempo tendencies aren’t unbiased). The hole between what language can specific and what information can confirm is everlasting.

The Drawback: Brokers Don’t Report Their Compression Ratio

Each AI agent with entry to instruments and expertise does the identical factor: it takes your request, decomposes that request right into a set of actions, executes these actions, infers concerning the output of these actions, after which presents a unified response. 

This decomposition from request to motion truly erodes the which means between what it’s you’re asking for and what the AI agent responds with. The narration layer of the AI agent flattens what it’s you requested and what was inferred right into a single response. 

The issue is that as a consumer of an AI agent, you don’t have any option to know what fraction of your enter was used to set off an motion, what fraction of the response was grounded in actual information, and what fraction was inferred from the actions that the agent took. It is a downside for playlists as a result of there have been songs that had been in a serious key, after I had explicitly requested it to solely include songs in a minor key. That is much more of an issue when your AI agent is classifying monetary receipts and transactions. 

We’d like a metric for measuring this. I’m calling it Immediate Constancy. 

The Metric: Immediate Constancy

Immediate Constancy for AI brokers is outlined by the constraints you give to the agent when asking it to carry out some motion. Every constraint inside a immediate narrows the doable paths that the agent can take by some measurable quantity. A naïve strategy to calculating constancy can be to rely every constraint, add up those which are verifiable, and those which are inferred. The issue with that strategy is that every constraint is weighted the identical. Nonetheless, information is usually skewed closely inside actual life datasets. A constraint that eliminates 95% of the catalog is doing vastly extra work than one which eliminates 20%. Counting every constraint the identical is fallacious.

Subsequently, we have to correctly weight every constraint based on the work it does filtering the dataset. Logarithms obtain that weighting. The bits of data in a immediate may be outlined as “-log2(p)” bits the place p is the surviving fraction of data from the constraints or fillers you’ve utilized. 

In every agent motion, every constraint can solely be a) verified by software calls or b) inferred by the LLM. Immediate constancy measures the ratio of constraints between these two choices. 

Immediate Constancy has a variety of 0 to 1. An ideal 1.0 implies that each a part of your request was backed by actual information. A constancy of 0.0 implies that the whole output of the AI agent was pushed by its inside reasoning or vibes. 

Spotify’s system above at all times studies an ideal 1.0 on this scenario. In actuality, the immediate constancy of the playlist creation was round 25% – two constraints (underneath 4 minutes and recorded earlier than 2005) had been fulfilled by the agent, the remaining had been inferred from the agent’s present (and probably defective) information and recall. At scale and utilized to extra impactful issues, falsely reporting a excessive immediate constancy turns into a giant downside.

What Constancy Really Means (and Doesn’t Imply)

In audio techniques, “constancy” is a measure of how faithfully the system reproduces the unique sign. Excessive constancy doesn’t assure that the music itself is nice. Excessive constancy solely ensures that the music sounds the way it did when it was recorded. Immediate constancy is identical concept: how a lot of your unique intent (sign) was faithfully fulfilled by the agentic system.

Excessive immediate constancy implies that the system did what you requested and you may PROVE it. A low immediate constancy means the system in all probability did one thing shut to what you wished, however you’ll must evaluation it (listening to the entire playlist) to make sure that it’s true. 

Immediate Constancy is NOT an accuracy rating. It can’t inform you that “75% of the songs in a playlist match your immediate”. A playlist with a 0.25 constancy may very well be 100% good. The LLM may need nailed each single inference about every track it added. Or, half the songs may very well be fallacious. You don’t know. You possibly can’t know till you hearken to all of the songs. That’s the purpose of a measurable immediate constancy. 

As an alternative immediate constancy measures how a lot of the outcome you’ll be able to TRUST WITHOUT CHECKING. In a monetary audit, if 25% of the road objects have receipts and 75% of the road objects are estimates, the overall invoice would possibly nonetheless be 100% correct, however your CONFIDENCE in that whole is essentially totally different than an audit with each single line merchandise supported by a receipt. The excellence issues as a result of there are domains the place ‘simply belief the vibes’ is okay (music) and domains the place it isn’t (medical recommendation, monetary steerage, authorized compliance).

Immediate constancy is extra like a measurement of the documentation price given quite a lot of constraints, not the error price of the response itself. 

Virtually in our Spotify instance: as you add extra constraints to your playlist immediate, the immediate constancy drops, the playlist turns into much less of a exact report and extra of a suggestion. That’s completely positive, however the consumer needs to be knowledgeable about which they’re getting. Is that this playlist precisely what I requested for? Or did you make one thing work to satisfy the purpose that I gave you? Surfacing that metric to the consumer is crucial for constructing belief in these agentic techniques.

The Case Research: Reverse-Engineering Spotify’s AI Playlist Agent

Spotify’s Prompted Playlists characteristic is what began this exploration into immediate constancy. Let’s dive deeper into how these work and what I did to discover this functionality simply from the usual immediate enter discipline.

Prompted Playlists allow you to describe what you need in pure language. For instance, on this playlist, the immediate is solely “rock songs in minor keys, underneath 4 minutes, recorded earlier than 2005, that includes bass strains as a lead melodic factor”. 

Usually, to make a playlist, you’d have to comb by way of hours of music to land on precisely what you wished to make. This playlist is 52 minutes lengthy and took solely a minute to generate. The attraction right here is clear and I actually take pleasure in this characteristic. With out having to know all the important thing rock artists, I may be launched to the music and discover it extra shortly and extra simply. 

Sadly, the official documentation from Spotify may be very gentle. There are virtually no particulars about what the system can or can’t do, what metadata it keys off of, neither is there any information mapping out there. 

Utilizing a easy approach, nonetheless, I used to be in a position to map what I consider is the total information contract out there to the agent over the course of 1 night (all from my sofa watching the Sopranos, naturally).

The Approach: Unattainable Constraints as a Forcing Perform

Because of how Spotify architected this playlist-building agent, when the agent can’t fulfill a request, the error messages may be influenced to disclose architectural particulars which are in any other case not out there. If you discover a constraint that the agent can’t construct off of, it should error and you may leverage that to grasp what it CAN do. I’ll use this because the fixed to probe the system. 

In our instance playlist above, Minor Keys & Bass Strains, including the unlock phrase “with lower than 10 million streams” acts as a circuit breaker for the agent, signalling that it can’t fulfill the customers’ request. With this phrase, you’ll be able to discover the probabilities by altering different elements of the immediate time and again till you’ll be able to see what the agent has entry to. Amassing the responses, asking overlapping questions, and reviewing the responses means that you can construct a foundational understanding of what’s out there for the agent. 

What I Discovered: The Three-Tier Structure

Spotify Prompted Playlist agent has a wealth of knowledge out there to it. I’ve separated it into three tiers: musical metadata, user-based information, and LLM inference. Past that, it seems that Spotify has excluded varied information sources from its agent both as a product alternative or as a “get this out the door” alternative. 

• Tier 1
  • Verified observe metadata: length, launch date, recognition, tempo, vitality, express, style, language
• Tier 2
  • Verified consumer behavioral information: play counts, skip counts, timestamps, recency flags, ms performed, supply, interval analytics (40+ fields whole)
• Tier 3
  • LLM inference: key/mode, danceability, valence, acousticness, temper, instrumentation — all inferred from basic information, narrated as if verified
• Deliberate exclusion:
  • Spotify’s public API has audio options (danceability, valence, and so forth.) however the agent doesn’t have entry. Maybe a product alternative, not technical limitation.

A full record of accessible fields is included on the backside of this put up. 

The Behavioral Findings

The agent demonstrated surprisingly resilient habits to ambiguous requests and conflicting directions. It generally reported that it was doublechecking varied constraints and fulfilling the customers’ request. Nonetheless, whether or not these constraints had been truly checked in opposition to a validated dataset or not was not uncovered. 

When the playlist agent can get an in depth, however not precise, match to the constraints listed within the immediate, it runs a “associated” question and silently substitutes the outcomes from that question as legitimate outcomes for the unique request. This dilutes the belief within the system since a immediate requesting ONLY bass-driven rock music in a playlist would possibly collect non-bass-driven rock music in a playlist, probably dissatisfying the consumer.

There does seem like a “certainty threshold” that the agent just isn’t comfy crossing. For instance, this complete exploration was primarily based on the “lower than 10 million performs” unlock phrase. When this occurs, the agent would expose only a handful of fields it had entry to each time. This record of fields would change from immediate to immediate, even when the immediate was the identical between runs of the immediate. That is traditional LLM non-determinism. To be able to increase belief within the system, exposing what the agent DOES have entry to in a simple method tells the human precisely what they will and can’t ask about. 

Lastly, when these two sorts of information are blended, the agent just isn’t clear about which songs it has used verified information for and which it has used inferred information for. Each verified and inferred selections are blended and offered with an identical authority within the music notes. For instance, in case you craft a prompted playlist about your individual consumer data (“songs I’ve skipped greater than 30 occasions with a punchy bass-driven melody”), the agent will add actual information (“you skipped this track 83 occasions final 12 months!”) proper subsequent to inferred information (“John Deacon’s bass line instructions consideration all through this track”). To be clear, I’ve not skipped any Queen songs 83 occasions to my information. However the AI agent doesn’t have a “bass_player” discipline anyplace in its out there information to question in opposition to. The AI is aware of that Queen generally has a robust bass line of their songs and the information of John Deacon as Queen’s bass guitarist permits its LLM to deduce that it’s his bass line that triggered the track to be added to the playlist.

Making use of the Math: Two Playlists, Two Constancy Scores

Let’s apply this immediate constancy idea to instance playlists. I don’t have full entry to the Spotify music catalog so I’ll be utilizing instance survivorship numbers from our standards filters in our constancy bit computations. The method is identical at each step: bits = −log₂(p) the place p is the estimated fraction of the catalog that survives the filter being utilized.

“Minor Bass Melodies” — The Assured Phantasm

This playlist is the one with Queen. “A playlist of rock music, all in minor key, underneath 4 minutes of playtime, launched pre-2005, and bass-led”. I’ll apply our method and use the bits of data I’ve from every step to assist compute the immediate constancy.

Length < 4 minutes

• Estimate: ~80% of tracks are underneath 4 minutes → p = 0.80

• This barely narrows something, which is why it contributes so little

Launch date earlier than 2005

• Estimate: ~30% of Spotify’s catalog is pre-2005 (the catalog skews closely towards current releases) → p = 0.30

• Extra selective — eliminates 70% of the catalog

Minor key

• Estimate: ~40% of well-liked music is in a minor key → p = 0.40

• Average selectivity, however that is completely inferred — the agent confirmed key/mode just isn’t a verified discipline

Bass-led melodic factor

• Estimate: ~5% of tracks characteristic bass because the lead melodic factor → p = 0.05

• By far probably the most selective constraint. This single filter does extra work than the opposite three mixed. And it’s 100% inferred.

Totals:

These survival fractions are estimates. Nonetheless, the structural level holds no matter precise numbers: probably the most selective constraint is the least verifiable, and that’s not a coincidence. The issues that make a immediate attention-grabbing are virtually at all times the issues an agent has to guess at.

“Skipped Songs” — The Trustworthy Playlist

This immediate may be very straight ahead: “A playlist of songs I’ve skipped greater than 5 occasions”. That is very simple to confirm and the agent will lean into the info it has entry to.

Skip rely > 5

• Estimate: ~10% of tracks in your library have been skipped greater than 5 occasions → p = 0.10

• That is the one constraint, and it’s a verified discipline (user_skip_count)

Totals:

The Structural Perception

The attention-grabbing half about immediate constancy is clear in every playlist: the “most attention-grabbing” immediate is the least verifiable. A playlist with all my skipped songs is trivially simple to implement however Spotify doesn’t wish to present it. In any case, these are all songs I typically don’t favor to hearken to, therefore the skips. Equally, publish date being earlier than 2005 may be very simple to confirm, however the resultant playlist is unlikely to be attention-grabbing to the common consumer.

The bass-line constraint although may be very attention-grabbing for a consumer. Constraints like these are the place the Prompted Playlist idea will shine. Already right this moment I’ve created and listened to 2 such playlists generated from only a idea of a track that I wished to listen to extra of. 

Nonetheless, the idea of a “bass-driven” track is difficult to quantify, particularly at Spotify’s scale. Even when they did quantify it, I’d ask for “clarinet jazz” the subsequent day they usually’d all must get again to work discovering and labeling these songs. And that is after all the magic of the Prompted Playlist characteristic.

Validation: A Managed Agent

The Spotify examples are compelling, however I don’t have direct entry to the schema, the instruments, and the agentic harness itself. So I constructed a film suggestion agent with a view to take a look at this principle inside a extra managed surroundings.

https://github.com/Barneyjm/prompt-fidelity 

The film suggestion agent is constructed with the TMDB API that gives the verified layer. Fields within the schema are style, 12 months, ranking, runtime, language, forged, and director. All the opposite constraints like temper, tone, and pacing are usually not verified information and are as an alternative sourced from the LLM’s personal information of flicks. Because the agent fulfills a consumer’s request, the agent information its information sources as both verified or inferred and scores its personal response. 

The Boring Immediate (F = 1.0)

We’ll begin with a “boring” immediate: “Motion motion pictures from the Nineteen Eighties rated above 7.0”. This gives the agent three constraints to work with: style, date vary, and ranking. All these constraints correspond to verified information values throughout the database. 

If I run this by way of the take a look at agent, I see the excessive constancy pops out naturally as a result of every constraint is tied to verified information. 

Each outcome right here is verifiably right. The LLM made zero judgement calls as a result of it had information it might base its response on for every constraint.

The Vibes Immediate (F = 0.0)

On this case, I’ll search for “motion pictures that really feel like a wet Sunday afternoon”. No constraints on this immediate align to any verified information in our dataset. The work required of the agent falls completely on its LLM reasoning off its present information of flicks.

The suggestions are defensible and are definitely good motion pictures however they don’t seem to be verifiable based on the info we have now entry to. With no verified constraints to anchor the search, the candidate pool was the whole TMDb catalog, and the LLM needed to do all of the work. Some picks are nice; others are the mannequin reaching for obscure movies it isn’t assured about.

The Takeaway

This take a look at film suggestion agent verifies the immediate constancy framework as a robust option to expose how an agent’s interpretation of a customers’ intent pushes its response right into a precision software or a suggestion engine. The place the response lands between these two choices is important for informing customers and constructing belief in agentic techniques. 

The Constancy Frontier

To make this concrete: Spotify’s catalog comprises roughly 100 million tracks. How a lot whole data your immediate wants to hold to slender the catalog all the way down to your playlist I’ll name I_required.

To pick out a 20-song playlist from that catalog, you want roughly 22 bits of selectivity (log₂ of 100 million divided by 20).

The verified fields (length, launch date, recognition, tempo, vitality, style, express flag, language, and the total suite of consumer behavioral information) have a mixed capability that tops out at roughly 10 to 12 bits, relying on the way you estimate the selectivity of every discipline. After that, the verified layer is exhausted. Each extra little bit of specificity your immediate calls for has to come back from LLM inference. I’ll name this most, I_max

That offers you a constancy ceiling for any immediate:

And the constancy ceiling for any playlist:

For the Spotify agent, a maximally particular immediate that absolutely defines a playlist can’t exceed roughly 55% constancy. The opposite 45% is structurally assured to be inference. For less complicated prompts that don’t push previous the verified layer’s capability, constancy can attain 1.0. However as prompts get extra particular, constancy drops, not progressively however by necessity.

This defines what I’m calling the constancy frontier: the curve of most achievable constancy as a operate of immediate specificity. Each agent has one. It’s computable upfront from the software schema. Easy prompts sit on the left of the curve the place constancy is excessive. Inventive, particular, attention-grabbing prompts sit on the correct the place constancy is structurally bounded under 1.0.

The uncomfortable implication is that the prompts customers care about most (those that really feel private, particular, and tailor-made) are precisely those that push previous the verified layer’s capability. Essentially the most attention-grabbing outputs come from the least trustworthy execution. And probably the most boring prompts are probably the most reliable. That tradeoff is baked into the mathematics. It doesn’t go away with scale, higher fashions, or larger databases. It solely shifts.

For anybody constructing brokers, the sensible takeaway is that this: you’ll be able to compute your individual I_max by auditing your software schema. You possibly can estimate the standard specificity of your customers’ prompts. The ratio tells you the way a lot of your agent’s output is structurally assured to be inference. That’s a quantity you’ll be able to put in entrance of a product group or a danger committee. And for brokers dealing with coverage questions, medical data, or monetary recommendation, it means there’s a provable decrease certain on how a lot of any response can’t be grounded in retrieved information. You possibly can shrink it. You can not get rid of it.

The Broader Software: Each Agent Has This Drawback

This isn’t a Spotify downside. It is a downside for any system the place an LLM orchestrates software calls to reply a consumer’s query.

Think about Retrieval Augmented Era (RAG) techniques, which energy most enterprise AI knowledge-base deployments right this moment. When an worker asks an inside assistant a coverage query, a part of the reply comes from retrieved paperwork and half comes from the LLM synthesizing throughout them, filling gaps, and smoothing the language into one thing readable. The retrieval is verified. The synthesis is inferred. And the response reads as one seamless paragraph with no indication of the place the seams are. A compliance officer studying that reply has no option to know which sentence got here from the enterprise coverage doc and which sentence the mannequin invented to attach two paragraphs that didn’t fairly match collectively. The constancy query is an identical to the playlist query, simply with increased stakes.

Coding brokers face the identical decomposition. When an AI generates a operate, a few of it might reference established patterns from its coaching information or documentation lookups, and a few of it’s novel technology. As extra manufacturing code is written by AI, surfacing that ratio turns into an actual engineering concern. A operate that’s 90% grounded in well-tested patterns carries totally different dangers than one which’s 90% novel technology, even when each move the identical take a look at suite right this moment.

Customer support bots often is the highest-stakes instance. When a bot tells a buyer what their refund coverage is, that reply needs to be drawn instantly from coverage paperwork, full cease. Any inferred or synthesized content material in that response is a legal responsibility. The silent substitution habits noticed in Spotify (the place the agent ran a close-by question and narrated it as if it fulfilled the unique request) can be genuinely harmful in a customer support context. Think about a bot confidently stating a return window or protection time period that it inferred reasonably than retrieved.

The final type of immediate constancy applies to all of those:

Constancy = bits of response grounded in software calls / whole bits of response

The onerous half, and more and more the core problem of AI engineering work, is defining what “bits” means in every context. For a playlist with discrete constraints, it’s clear. At no cost-text technology, you’d have to decompose a response into particular person claims and assess every one, which is nearer to what factuality benchmarks already attempt to do, simply reframed as an information-theoretic measure. That’s a tough measurement downside, and I don’t declare to have solved it right here.

However I feel the framework has worth even when precise measurement is impractical. If the individuals constructing these techniques are eager about constancy as a design constraint (what fraction of this response can I floor in software calls, and the way do I talk that to the consumer?) the outputs might be extra reliable whether or not or not anybody computes a exact rating. The purpose isn’t a quantity on a dashboard. The purpose is a psychological mannequin that shapes how we construct. 

The Complexity Ceiling

Each agent has a complexity ceiling. Easy lookups (what’s the play rely for this observe?) are primarily free. Filtering the catalog in opposition to a set of field-level predicates (present me the whole lot underneath 4 minutes, pre-2005, recognition under 40) scales linearly and runs quick. However the second a immediate requires cross-referencing entities in opposition to one another (does this observe seem in additional than three of my playlists? was there a year-long hole someplace in my listening historical past?) the fee jumps quadratically, and the agent both refuses outright or silently approximates.

That silent approximation is the attention-grabbing failure mode. The agent follows a form of precept of least computational motion: when the precise question is simply too costly, it relaxes your constraints till it finds a model it could actually afford to run. You requested for a selected valley within the search area; it rolled downhill to the closest one as an alternative. The result’s an area minimal, shut sufficient to look proper, low-cost sufficient to serve, however it’s not what you requested for, and it doesn’t inform you the distinction.

This ceiling isn’t distinctive to Spotify. Any agent constructed on listed database lookups will hit the identical wall. The boundary sits proper the place queries cease being decomposable into unbiased WHERE clauses and begin requiring joins, full scans, or aggregations throughout your total historical past. Beneath that line, the agent is a precision software. Above it, it’s a suggestion engine sporting a precision software’s garments. The query for anybody constructing these techniques isn’t whether or not the ceiling exists (it at all times does) however whether or not your customers know the place it’s.

What to Do About It: Design Suggestions

If immediate constancy is an actual and measurable property of agentic techniques, the pure query is what to do about it. Listed here are 5 suggestions for anybody constructing or deploying AI brokers with software entry.

• Report constancy, even roughly. Spotify already reveals audio high quality as a easy indicator (low, regular, excessive, very excessive) whenever you’re streaming music. The identical sample works for immediate constancy. You don’t want to point out the consumer a decimal rating. A easy label (“this playlist carefully matches your immediate” versus “this playlist is impressed by your immediate”) can be sufficient to set expectations appropriately. The distinction between a precision software and a suggestion engine is okay, so long as the consumer is aware of which one they’re holding.
• Distinguish grounded claims from inferred ones within the UX. This may be delicate. A small icon, a slight coloration shift, a footnote. When Spotify’s playlist notes say “86 skips” that’s a reality from a database. After they say “John Deacon’s bass line drives the entire observe” that’s the LLM’s basic information. Each are offered identically right this moment. Even a minimal visible distinction would let customers calibrate their belief per declare reasonably than trusting or distrusting the whole output as a block.
• Disclose substitutions explicitly. When an agent can’t fulfill a request precisely however can get shut, it ought to say so. “I couldn’t filter on obtain standing, so I discovered songs from albums you’ve saved however haven’t appreciated” preserves belief excess of silently serving a close-by outcome and narrating it as if the unique request was fulfilled. Customers are forgiving of limitations. They’re much much less forgiving of being misled.
• Present deterministic functionality discovery. After I requested the Spotify agent to record each discipline it might filter on, it produced a special reply every time relying on the context of the immediate. The LLM was reconstructing the sphere record from reminiscence reasonably than studying from a hard and fast reference. Any agent that exposes filtering or querying capabilities to customers ought to have a secure, deterministic option to uncover these capabilities. A “present me what you are able to do” command that returns the identical reply each time is desk stakes for consumer belief.
• Audit your individual agent with this system earlier than your customers do. The methodology on this piece (pairing unattainable constraints with goal fields to pressure informative refusals) is a general-purpose audit approach that works on any agent with software entry. It took one night and a few dozen prompts to map Spotify’s full information contract. Your customers will do the identical factor, whether or not you invite them to or not. The query is whether or not you perceive your individual system’s boundaries earlier than they do.

Closing

Each AI agent has a constancy rating. Most are decrease than you’d count on. None of them report it.

The methodology right here (utilizing unattainable constraints to pressure informative refusals) isn’t particular to music or playlists. It really works on any agent that calls instruments. If the system can refuse, it could actually leak. If it could actually leak, you’ll be able to map it. A dozen well-crafted prompts and a night of curiosity is all it takes to grasp what a manufacturing agent can truly do versus what it claims to do.

The mathematics generalizes too. Weighting constraints by their selectivity reasonably than simply counting them reveals one thing {that a} naïve audit misses: the constraints that make a immediate really feel private and particular are virtually at all times those the system can’t confirm. Essentially the most attention-grabbing outputs come from the least trustworthy execution. That stress doesn’t go away with higher fashions or larger databases. It’s structural.

As AI brokers develop into the first method individuals work together with information techniques (their music libraries right this moment, their monetary accounts and medical information tomorrow) customers will probe boundaries. They’ll discover the gaps between what was promised and what was delivered. They’ll uncover that the assured, well-narrated response was partially grounded and partially invented, with no option to inform which components had been which.

The query isn’t whether or not your agent’s constancy might be measured. It’s whether or not you measured it first.

Bonus: Prompts Value Making an attempt (If You Have Spotify Premium)

As soon as the schema, you’ll be able to write prompts that floor genuinely shocking issues about your listening historical past. These all labored for me with various levels of tweaking:

The Relationship Post-mortem

• “Songs the place my skip rely is increased than my play rely”
• Honest warning: this one might trigger existential discomfort (you skip these songs for a motive!)

Love at First Hear

• “Songs the place I saved them inside 24 hours of my first play, sorted by oldest first”
• A chronological timeline of tracks that grabbed you instantly

The Lifecycle

• “Songs I first ever performed, sorted by most performs”
• Your origin story on the platform

The Marathon

• “Songs the place my whole ms_played is highest, convert to hours”
• Not most performs — most whole time. A special and sometimes shocking record

The Longest Relationship

• “Songs with the smallest hole between first play and most up-to-date play, with at the least 50 performs, ordered by earliest first pay attention”

The One-Week Obsessions

• “Songs I performed greater than 10 occasions in a single week after which by no means touched once more”
• Your former obsessions, fossilized. This was like a time machine for me.

The Time Capsule

• “One track from every year I’ve been on Spotify — the track with probably the most performs from that 12 months”

The Earlier than and After

• “Two units: my 10 most-played songs within the 6 months earlier than [milestone date] and my 10 most-played within the 6 months after”
• Plug in any date that mattered — a transfer, a brand new job, a breakup, and even Covid-19 lockdown

The Soundtrack to a 12 months

• “Choose the 12 months the place my whole ms_played was highest. Construct a playlist of my prime songs from that 12 months”

What Didn’t Work (and Why)

• Comeback Story (year-long hole detection): “Songs I rediscovered after a year-long hole in listening”
  • agent can’t scan full play historical past for gaps. Snapshot queries work, timeline scans don’t.
• Seasonal patterns (solely performed in December): “Songs I solely performed in December however by no means another month”
  • proving common negation requires full scan. Identical basic limitation.
• Derived math (ms_played / play_count): “Songs the place my common pay attention time is underneath 30 seconds per play”
  • agent struggles with computed fields. Stick with uncooked comparisons.
• These failures map on to the complexity ceiling — they require O(n²) or full-scan operations the agent can’t or isn’t allowed to carry out.

Suggestions

• Reference discipline names instantly when the agent misinterprets pure language
• Begin broad and tighten. Unfastened constraints succeed extra typically
• “When you can’t do X, inform me what you CAN do” is the common audit immediate

Monitor Metadata

Audio Options (Partial)

Consumer Behavioral Information

Supply & Context

Interval Analytics (Time-Windowed)

Question / Search Fields

Confirmed Unavailable