INTENT AI
Capability, vision & roadmap · Confidential
Capability brief
INTENT AI

The Human Context Layer for the agent era.

Predicting the moment each individual is ready to act — fusing a brand's own data with 0-party signal from the device — at the scale of hundreds of millions of customers.

What it is, how it works, and what it's worth, a brief for the people who build it and the people who sell it.
The opportunity

The opportunity is to stop missing the moment.

The same engine reads three very different lives. Each one is forming a moment of need in their behaviour, weeks before they go looking, and weeks before they tell a single brand.

Amara is about to move house. She hasn't told a single brand, and may not have fully decided herself. Intent AI sees it forming weeks early, and surfaces the moment to offer broadband, a family plan and the right financing, before she goes looking.

Routines
Weekend browsing drifts to a new neighbourhood
Topics & Interests
Mortgage rates, school catchments
Patterns
Listings, then financing, the pre-move sequence
Intents
Comparing home broadband
Moment of need
About to move within 6 weeks: broadband, family plan, financing

Daniel already banks with you — a business account, two years in. Now the business is scaling: first hires on payroll, suppliers abroad, card volume climbing, a cash-flow gap opening. Intent AI reads the growth weeks early and surfaces the moment to upsell — lending, FX, payroll and a business card — before he shops around. One account becomes the whole relationship.

Routines
Payroll runs, daily card settlement
Topics & Interests
Business lending, FX, merchant services
Patterns
Hiring, then overseas suppliers, then a cash gap
Intents
Financing the next stage of growth
Moment of need
Ready to upsell: lending, FX, payroll, business card

Leah is drifting away, and the brand can't see it yet, a loyal weekly shopper now quietly slowing down. Intent AI catches it while there's still time to act, with a reward or a reason to stay, before she defaults to a competitor. A proactive win-back beats a discount sent after she's gone.

Routines
Visits slow, baskets shrink, late-night browsing
Topics & Interests
Competitor apps, voucher sites, sale items
Patterns
Fewer visits, then rival activity, then idle points
Intents
Looking for a better price elsewhere
Moment of need
About to lapse: personalised reward, category nudge, tier-save

Consumers now ask an AI agent to find the best mortgage, the best broadband, the best deal, and the brand never gets a look in. Human context surfaces the moment of need before the consumer turns to AI, giving brands the chance to act first.

Events & sequences

Behaviour is a language. We read it in sequence.

A person's life is a string of events laid out on one axis — time. Each event is a single token built from four dimensions — when, where, what (the brand they touched) and who — and each dimension is two layers: the 1st-party data a brand already holds, and a 0-party layer from Intent's on-device SDK, signal the brand could get no other way. Topics and interests aren't a channel; they're read off the brand. Time is the spine, tokens are the words, sequences are the sentences — read them in order and the next move becomes predictable.

ANATOMY OF ONE EVENT FOUR DIMENSIONS · 1P BRAND DATA + 0P INTENT SDK (DEVICE) When TIME · DURATION 1P · BRAND DATA Sat, late AM 0P · INTENT SDK long dwell · weekend rhythm Where PLACE · CONTEXT 1P · BRAND DATA postal area · billing region 0P · INTENT SDK home geo · weekend at-home What BRAND · INTERACTION 1P · BRAND DATA Rightmove → Housing 0P · INTENT SDK property app · listings screen Who PROFILE · IDENTITY 1P · BRAND DATA 4-yr customer · family plan · high value 0P · INTENT SDK device rhythm · companion brands ONE EVENT · AT t Rightmove TIME · THE AXIS EVERY EVENT SITS ON ← past future → t 1st-party · brand data 0-party · Intent Edge SDK (device)

One weekend in Amara's data — tap any event to load it into the token above.

Sat · late AMRightmoveHousing
Sat · late AMZooplaHousing
Sun · PMMortgage rate comparisonFinance
Sun · PMSchool catchment lookupFamily
Mon · eveningBroadband plansTelco
ANATOMY OF ONE EVENT FOUR DIMENSIONS · 1P BRAND DATA + 0P INTENT SDK (DEVICE) When TIME · DURATION 1P · BRAND DATA Month-end 0P · INTENT SDK recurring · fixed cadence Where PLACE · CONTEXT 1P · BRAND DATA business address · region 0P · INTENT SDK office geo · merchant locations What BRAND · INTERACTION 1P · BRAND DATA Payroll → Salaries 0P · INTENT SDK banking app · payroll screen Who PROFILE · IDENTITY 1P · BRAND DATA 2-yr business · multi-product 0P · INTENT SDK app-usage rhythm · finance apps ONE EVENT · AT t Payroll TIME · THE AXIS EVERY EVENT SITS ON ← past future → t 1st-party · brand data 0-party · Intent Edge SDK (device)

One month in Daniel's data — tap any event to load it into the token above.

Month-endPayroll runPayroll
TueInternational paymentFX
DailyCard settlement upPayments
Late monthOverdraft touchedLiquidity
EveningBusiness lending searchCredit
ANATOMY OF ONE EVENT FOUR DIMENSIONS · 1P BRAND DATA + 0P INTENT SDK (DEVICE) When TIME · DURATION 1P · BRAND DATA Late night, weekend 0P · INTENT SDK late-night · long dwell Where PLACE · CONTEXT 1P · BRAND DATA delivery region 0P · INTENT SDK home geo · late-night, on sofa What BRAND · INTERACTION 1P · BRAND DATA ASOS → Fashion 0P · INTENT SDK rival app · browse screen Who PROFILE · IDENTITY 1P · BRAND DATA high lifetime spend · loyalty tier 0P · INTENT SDK declining engagement · rival apps ONE EVENT · AT t ASOS TIME · THE AXIS EVERY EVENT SITS ON ← past future → t 1st-party · brand data 0-party · Intent Edge SDK (device)

One week in Leah's data — tap any event to load it into the token above.

SatASOSFashion · rival
SunVoucherCodesDiscounts
TueAbandoned basketOwn brand
ThuBrowses sale items onlyOwn brand
OngoingPoints sit unusedLoyalty

The sequence is the moat. Read in order, the next event is predicted and checked against what actually happened — provable, not a guess. A few integers per prediction, not millions of tokens; the patterns learned across trillions of tokens are the edge, not the data.

Human context

Everything we understand about one person.

One event is evidence; a sequence is behaviour. Put enough of it together and a whole individual comes into view — not a segment, a person; in How, this same picture is the world model's view of one life. Here is Amara, and the parts of her life Intent AI reads, all pointing to the one moment a business can act on.

From when

Routines

Weekend hours drifting toward a new neighbourhood; commute-route lookups.

From what

Topics & Interests

Property portals, mortgage rates, school catchments, broadband.

From who

Traits

Growing family, long tenure, high value, financially active.

Amara
Four-year customer
About to move house — and she hasn't told a soul.
From the sequence

Patterns

The pre-move sequence: listings, then financing, then home services.

From the sequence

Intents

Micro: browse listings. Macro: buy a home. Long-term: settle the family.

The prediction

Moment of need

About to set up a new home: broadband, a family plan, financing, insurance.

Every part of this traces back to events that actually happened, which is why the moment of need is a prediction you can check, not a story you tell.

What Intent AI predicts

Two predictions decide where to act.

For each individual, at a moment in time, Intent AI predicts both the likelihood of reaching a moment of need and the lift from acting on it. The second is where the value lives. Predicting who acts anyway earns little; predicting whom an action moves is the whole point.

Propensity · the moment of need
84% likely to move home within 8 weeks

Detected weeks before Amara lists or applies. A home move is the trigger for a chain of needs: broadband, a family plan, financing, insurance.

Persuadability · causal uplift
+31 pts uplift on the right offer, at the right moment

A lookalike in her segment moves only +5 points. Act on the individual at her moment of need, not the average. Persuadability, not raw propensity, is the north star. In How, this number is the simulator's read — the lift it predicts before a penny is spent.

Together they define an opportunity: a specific person, at a specific moment, where acting actually changes the outcome. The opportunity — not the segment — is the unit a campaign acts on.

From opportunity to action

Consumers got agents. Brands need their own.

Predicting the opportunity is only half of it — someone has to act, fast, while the moment is open. Consumers already have AI agents acting for them; the answer for brands is an agentic layer of their own. You set the goal; Intent surfaces the opportunities, simulates the options, and acts — then the response flows back and sharpens the next call.

You · the goal

Goals & preferences

Tell Intent the outcome — grow broadband, win back lapsers, deepen the relationship. No audience-building, no rules to hand-write.

Intent · the read

Opportunity intelligence

Every individual at a real moment of need, ranked by persuadability. Opportunities, not segments — the unit you act on.

Agents · the act

Autonomous marketing

Campaign agents rehearse each option in simulation and act only on the ones that move people — inside your guardrails.

Goal in, opportunities out, action taken, response learned. One closed loop — and in How, it is the same loop the architecture runs end to end: data, cognition, agency, and back to data.

Why it wins

Valuable, cheap, and provable, all at once.

A prediction only counts if it produces value, costs little to make, and can be proven right. Most platforms win one and lose the others. The discipline is holding all three.

Valuable
ROI

Measured as lift over a hold-out control group. Proven at 15 to 34× across the portfolio.

Cheap
Cost / prediction

The lowest-cost inferencing in the category, driven by a global pattern library that means we never relearn the world.

Provable
Accuracy

Validated against held-out reality. A prediction you can't check is an opinion, not an asset.

The proof

In production with hundreds of millions of customers.

300M+
customer profiles processed monthly
$150M+
measured revenue uplift
15–34×
return on investment
7 yrs
Verizon, 100M customers, renewed

For telco and financial services, the only platform handling this at agentic scale without hallucinating. Partners include Verizon, MTN and Grab.

Why it's defensible

Governed knowledge, reused across millions.

The 0-party edge

On-device SDK

A 0-party signal the brand can't get any other way. Intent's Edge SDK enriches every channel of every event — sharpening rich first-party data, and standing in where it's thin. Privacy built in; the same SDK deploys across verticals — cyber, insurance, health, identity.

Governance

It thinks before it speaks

Confidence, provenance and safety on every prediction before it leaves. Agentic scale, without hallucination.

The moat

Global Pattern Library

Write once, read millions. Knowledge learned once is reused across every profile, driving accuracy up and inference cost far below the alternatives.

Stated honestly: models are client-scoped and we do not own the underlying customer data. The moat is the architecture, the methodology and the governed universal knowledge, not a single cross-client brain.

See How — the architecture behind all of this →

The vision · how it works

Behaviour is a language.
We build the world model that speaks it.

Tokenise every behavioural moment. Lay the tokens on a time axis. Learn the sequence at scale, self-supervised. Predict the next move. Simulate every move; act only when it causally matters.

Time is the spine. Tokens are the words. Sequences are the sentences. The world model is the reader.
Unit 01 · the behavioural token

Every moment becomes a behavioural token.

A behavioural token is one moment in a person's life, reduced to a few integers. Four dimensions define it: when (time), where (place), what (the brand they touched) and who (the user). Each of those four is built in two layers: the 1st-party data our customers already have, and a 0-party enrichment layer brought in by Intent's Edge SDK from the user's device — signal the brand could not get any other way. Time is also the spine: every token sits at a point on a timeline, and the model learns by reading those points in order.

ANATOMY OF A BEHAVIOURAL TOKEN FOUR DIMENSIONS · 1P CUSTOMER DATA + 0P INTENT SDK ENRICHMENT When TIME · DURATION 1P · CUSTOMER DATA timestamp · session start · day-of-week 0P · INTENT SDK dwell · inter-event interval · rhythm Where PLACE · CONTEXT 1P · CUSTOMER DATA postal area · billing region · channel 0P · INTENT SDK precise geo · vicinity · place-type What BRAND · INTERACTION 1P · CUSTOMER DATA brand touched · product · campaign 0P · INTENT SDK app · screen · category exposure Who PROFILE · IDENTITY 1P · CUSTOMER DATA tenure · plan · value · products held 0P · INTENT SDK on-device rhythm · habits · companion brands Behavioural token ONE MOMENT · AT t TIME · THE PRIMARY AXIS THAT EVERY TOKEN SITS ON ← past future → t 1st-party · customer data 0-party · Intent Edge SDK enrichment
When · time

Time & duration

1P: timestamp, day-of-week, session start. 0P · SDK: dwell, inter-event intervals, on-device rhythm.

Where · place

Place & context

1P: postal area, billing region, channel. 0P · SDK: precise geo, vicinity, type-of-place.

What · brand

Brand & interaction

1P: brand touched, product, campaign. 0P · SDK: app, screen, surrounding category exposure.

Who · profile

Identity & traits

1P: tenure, plan, value, products held. 0P · SDK: on-device rhythm, habits, companion brands.

Intent's edge SDK

A 0-party source that enriches every channel — not another input to it.

Every channel of the token is built in two layers: 1P data the customer gives us, and a 0P layer from Intent's Edge SDK on the user's device. The SDK does not sit alongside the four channels as a fifth signal — it sits across them, contributing consented, first-person evidence to when, where, what and who, to whatever extent the brand's own data leaves unsaid. That second layer is data the brand could not obtain any other way.

Whentime
Whereplace
Whatbrand
Whoprofile
Brand has rich data

SDK signal sharpens what's already there — device rhythm, real-world location and dwell, screen context the CRM never sees.

Brand has thin data

SDK signal stands in for the channels themselves — 0-party identity, place, time and intent from the device, so the model can perceive a user the brand barely knows.

Time is the spine. Four dimensions make the token. Each is a stack of 1P + 0P. The brand's first-party data lays the foundation; Intent's Edge SDK supplies the 0-party signal that would otherwise be missing. The model learns by reading those tokens in order, one slice of life at a time.

Unit 02 · the sequence

A life becomes a behavioural timeline.

Time is what makes this work. Concatenate tokens chronologically and a day, a week, a life unfolds as a single sequence. The model learns to read it like a language — Sleep, Instagram, Trainline, IntentHQ, Pret — because a life, like a sentence, has an order. Strip out the time order and you have a bag of moments with no meaning; keep it and you can predict the next move before it happens.

Behavioural timeline · a day
SleepAt Home
Social Media[Instagram]
Unknown
Travel[Trainline]
Work[IntentHQ]
Lunch[Pret]
Work[IntentHQ]
Socialising[Pub]
Travel[Trainline]
Unknown
Entertainment[Netflix]
SleepAt Home
The promise · predict the next behaviour
Sleep
Social Media[Instagram]
Unknown
Travel[Trainline]
Work[IntentHQ]
Predicted next · probability distribution
Lunch [Pret]
0.60
Lunch [Farmer J]
0.30
Break [Costa]
0.10
Pprior · from the world model
P( next | history )
Lunch [Pret]0.60
Lunch [Farmer J]0.30
Break [Costa]0.10
Simulator
Poutcome = f( Pprior, action )
The simulator takes the prior distribution and a candidate action — here, Farmer J sends 10% off now — and emits what the outcome distribution would look like under that action.
Poutcome · from the simulator
P( next | history, do(action) )
Lunch [Pret]0.20
Lunch [Farmer J]0.70
Break [Costa]0.10

Same sequence, two futures. The world model doesn't just predict what's likely — it answers what would change if a brand acted right now. That's the difference between targeting and incrementality.

Training paradigm · self-supervised

Sequence-to-sequence modelling at scale.

We never need a label — time supplies the supervision. At each point on the timeline the model sees the past and is asked to predict the next behavioural token. A causal mask enforces the arrow of time: the model can only attend to what has already happened — no leakage from the future. Run this across hundreds of millions of timelines and the model learns how lives actually unfold, moment by moment.

Next-token prediction · causally masked

Causal mask
Observed (past) Predict (next behaviour)
Objective

Next-token prediction

Cross-entropy on the next behavioural token. The same objective that powers language models, applied to the language of behaviour.

Signal

Self-supervised

No human labels. Every observed sequence supervises itself — the future of the sequence is its own target. Data scales without annotation cost.

Result

A world of behaviour

From trillions of tokens, the model internalises routines, rhythms, life-stages, and the conditions that change them.

Architecture · the world model

A world model of the user, not the web.

Following LeCun's autonomous-intelligence blueprint, we lift the architecture out of robotics and onto the person. Heterogeneous signals — edge, web, CRM — collapse into a single latent state s[t], but the load-bearing word is t: the state is a slice of a continuously evolving user. The predictor rolls it forward — s[t+1] = Pred(s[t], a[t], z) — emitting a prior distribution over the next behavioural token. The actor — our simulator — takes that prior together with a candidate action and emits the outcome distribution under that action. The cost module — taught by Causal AI — scores it. Every box in the diagram below is anchored to a point on the time axis at the top.

TIME · THE STATE EVOLVES ALONG THIS AXIS s[t-1] s[t] NOW · THIS DIAGRAM s[t+1] s[t+k] time → The User latent intent · context THE WORLD Enc_edge device · location · rhythm Enc_web browse · search · dwell Enc_crm tenure · products · value PERCEPTION · JEPA ENCODERS s[t] latent user state hierarchical · multi-horizon STATE Memory profile vector store Predictor H-JEPA · world model s[t+1] = Pred(s[t], a[t], z) short + long horizon PREDICTOR Cost C(s) Intrinsic · guardrails Critic · long-term value Causal AI · teacher immutable + learned + taught Causal AI TEACHER · CATE TEACHES THE COST & THE ACTOR Actor simulator P_outcome = f(P_prior, a) picks a · minimises C taught by Causal AI SIMULATOR P_prior SIMULATION · ACTION a[t] z UNCERTAINTY
Perception

JEPA encoders

One encoder per modality — edge, web, CRM. They discard the noise (the exact URL, the precise ms) and keep what is predictive of intent.

State

Latent s[t]

One hierarchical representation per user. Lower levels for short-horizon prediction, higher levels for life-stage and LTV.

Memory

Profile store

Streaming-updated profile vectors. Only what an event changes is touched — designed for billions of events a day.

Predictor

H-JEPA world model

Rolls the state forward: s[t+1] = Pred(s[t], a[t], z). A latent z captures the irreducible uncertainty of human behaviour.

Cost

Intrinsic + Critic + Causal AI

Immutable guardrails (privacy, frequency caps, fairness), a trained critic that scores long-term LTV and retention, and a Causal AI teacher term that distils CATE-uplift from past randomised holdouts directly into the cost signal driving the actor.

Actor

The simulator

Searches the space of content × channel × timing × offer, simulates each through the world model, picks the action that minimises expected future cost.

The simulator · causal AI

Act only when simulation says it causally moves the needle.

A propensity score answers "is this person likely to act?" — a question every competitor can answer. The simulator answers a harder one: "would this person act differently because of what we did?" The simulator is, concretely, a learned function that takes the world model's prior distribution and a candidate action, and emits the outcome distribution under that action.

The simulator
Poutcome = fθPprioraction )
Input · PpriorThe world model's distribution over the next behavioural token, given history.
Input · actionA candidate action: content × channel × timing × offer.
Output · PoutcomeThe shifted distribution the action is predicted to produce — the counterfactual world.
Causal uplift  =  Poutcome − Pprior  on the target outcome
Counterfactual reasoning

P(outcome | do(action), s[t])

The world model simulates both worlds — "no action" and "this action" — and compares the predicted trajectories. The gap is the causal uplift, the only quantity that matters for ROI.

Conditional Average Treatment Effect

CATE per individual

Trained with X-learner, DR-learner, causal forests and meta-learners on randomised holdouts and natural experiments. Per-individual treatment effect, not population averages.

Doubly-robust validation

Double Machine Learning

Separates the propensity (would they have acted?) from the outcome (did the action change anything?). Robust to confounders the encoders couldn't fully strip.

Exploration discipline

Randomised holdouts

A small, governed slice of impressions is randomised. That slice is the only ground truth for causality; everything else is observational and biased by the logging policy.

The decision the simulator actually makes

Causal AI sorts each individual into one of four cells. Only one is a real opportunity — and it is the only cell that deserves an impression. This is how we drive the cost of action down and the impact up.

High uplift
ACT NOW

Persuadables

Won't act on their own, will act if nudged. This cell is the opportunity — the only group where impressions create value, and where the simulator spends.

No uplift, won't convert
Skip

Lost Causes

Won't act regardless. The simulator saves the impression and the frequency-cap headroom for someone else.

Negative uplift
Do not disturb

Sleeping Dogs

Would act on their own but will react badly to contact. Targeting them destroys value. The simulator suppresses contact.

Teacher → Student

Causal AI is the teacher of the simulator

Source of truth

Past randomised holdouts

Thousands of controlled experiments. The only data where causality — not correlation — can be measured directly.

CATE · uplift labels
Teacher

Causal AI

X-learner, DR-learner, causal forests, double ML. Estimates the per-individual treatment effect that defines "what would actually change?"

distilled into cost
Student

Simulator

f(Pprior, action) → Poutcome. Scored against the teacher's uplift labels through the cost function, so the emitted Poutcome matches what would actually happen under that action.

Causal AI learns the truth slowly and expensively in controlled experiments. The simulator learns the truth quickly and cheaply by being taught — its loss is the teacher's CATE signal, baked into the world model's cost function. The result: causal decisions, at the speed of self-supervised serving.

The world model's cost function · taught by Causal AI
L_next-token + λc L_Causal-AI-teacher + λg L_guardrail

One unified loss for the world model and the simulator. Self-supervised next-token prediction trains the world model on the firehose. Causal AI acts as the teacher — its CATE-uplift estimates, learned from past randomised holdouts, become a supervised label that flows directly into the cost function and tells the simulator "did this action actually move the needle?". The immutable guardrail term sits inside the same cost so the actor cannot optimise it away.

The simulator's job is not to predict who acts — it is to predict whom an action changes. Causal AI is how the simulator earns the right to spend an impression, and how every action is rehearsed in simulation before it's served.

See it in motion

Autoregressive generation, then causal simulation.

Behaviour rolls forward like a language. The world model takes the sequence so far and emits a probability over the next behavioural token — pick the top, append, repeat. The simulator then asks the harder question: given everything we've learned from past randomised holdouts, which action would shift that distribution most?

Step 1 · world model

Autoregressive · predict the next behaviour

Behavioural sequence so far
World
model
Top-K next-token probabilities

The model emits a distribution over the next behavioural token. The argmax is selected, appended to the sequence, and becomes the next input. Generation rolls forward — exactly like an LLM, except the language is life.

Step 2 · simulator

f( Pprior, action )  =  Poutcome

Trained on past causal results · randomised holdouts
positive uplift null effect negative (sleeping dog)
Learned policy
For every candidate action, emit the outcome distribution and score the uplift over Pprior.
Pprior · from the world model (Step 1) f( Pprior, action ) → Poutcome
Lunch [Pret]
0.60
Lunch [Farmer J]
0.30
Break [Costa]
0.10
Each candidate action below shows the Poutcome the simulator would emit. Target outcome: Lunch [Farmer J].

Same prior, four candidate actions, four predicted outcome distributions. The uplift on the target outcome is the simulator's score for each action. Only the action with positive, significant uplift earns the impression.

The world model says what's likely. The simulator says what's worth doing. The first is autoregressive prediction; the second is causal inference grounded in every randomised holdout we've ever run.

Training & evaluation

Self-supervised at scale, causally validated.

Five stages, one closed loop, one shared axis: time. Raw behaviour is tokenised into behavioural tokens, per-modality encoders fuse the streams into a single time-ordered sequence over one vocabulary, a sequence model learns the patterns the way it would learn a language — by reading them in order — a causal-AI-distilled simulator scores what happens if we act, and every controlled market test feeds back into the next training pass. Behaviour evolves along the time axis; the model is trained to evolve with it.

FORWARD PIPELINE · PERCEIVE → PREDICT RAW SIGNAL 01 · TOKENISE 02 · ENCODE UNIFIED BEHAVIOURAL SEQUENCE · ONE SHARED VOCAB 03 · SEQ2SEQ WORLD MODEL Web logs CRM Edge data Market insights Tok_web Tok_crm Tok_edge Tok_market Enc_web Enc_crm Enc_edge Enc_market one tokeniser + one encoder per data modality single time-ordered sequence · shared vocab w₁ e₁ w₂ c₁ w₃ e₂ m₁ w₄ e₃ c₂ w₅ m₂ e₄ w₆ c₃ t₀ TIME · PRIMARY AXIS future → web crm edge market Seq2Seq World Model SELF-SUPERVISED · NEXT-TOKEN x₁ x₂ x₃ x₄ x₅ x_next? predicts next token in the shared vocab self-attention cross-modality fusion positional + temporal P(next | history, context) PRIOR PROBABILITIES priors 04 · INTERVENTION MODEL · ACTOR / CONTROLLER / SIMULATOR Candidate action a intervention to evaluate Prior P(next | ctx) from seq2seq f(prior, action) → P(outcome | do(a)) student model · distilled from causal-AI teacher Causal AI · Teacher DoWhy · DR-learner · CATE distillation Δ uplift = P(y|do(a)) − P(y) simulated counterfactual positive simulated uplift surfaces an opportunity 05 · CLOSED LOOP · MARKET TEST → REINFORCE Opportunity highest-uplift action Controlled test treatment vs. control Observed uplift real-world outcome Reinforce · retrain close the loop World Model · v_next continuous learning REINFORCEMENT SIGNAL Tokenisers Encoders Seq2Seq world model Intervention model Causal-AI teacher Closed-loop reinforcement
From raw modalities → context latents → next-token priors → counterfactual outcomes → market truth → reinforcement
01 · Tokenise
Behaviour → behavioural tokens

Before any model can read it, raw behaviour has to be made discrete. A per-modality tokeniser turns every behavioural moment — a page view, a CRM event, an edge ping, a market signal — into a behavioural token. One vocabulary per modality; the same language for every downstream stage.

Tok_web · Tok_crm · Tok_edge · Tok_mkt
02 · Encode
Per-modality encoders → one unified sequence

To perceive the user's context, each token stream goes through its own encoder. Every encoder projects its modality's tokens into the shared latent space. The streams then interleave by time into a single behavioural sequence over one shared vocabulary — the substrate the world model actually learns from.

Shared vocab · time-ordered · multi-source
03 · Predict
Sequence-to-sequence world model

A seq2seq model learns the patterns in those latent sequences. Trained self-supervised with a next behavioural token prediction objective — no labels, just the firehose. Its output is the prior probability over what the user is likely to do next.

Self-supervised · next-token
04 · Intervene
Intervention model · actor · simulator

An intervention model — the actor / controller / simulator — takes the prior probabilities together with a candidate action and produces the probabilities if that intervention were applied. Trained by distillation from a causal-AI teacher: counterfactual reasoning, baked into a fast model.

Causal-AI teacher · distillation
05 · Close the loop
Market test → continuous learning

The simulator surfaces an opportunity worth pursuing in the market. A controlled-group test gives the ground truth, and the result feeds back to reinforce learning. Every loop tightens the model — the path by which we get to the world model we envision.

Controlled test · reinforce · world model
Offline · perplexity

Next-token accuracy

How well does the world model predict the next behavioural token on held-out sequences? The primary loss, the cleanest signal of perception + prediction working together.

Causal · uplift AUC

Qini & Uplift curves

How well does the simulator rank persuadables above sure-things and sleeping dogs? Measured against the causal-AI teacher and against randomised holdouts — not the logging policy.

Online · A/B

Lift vs. control

The final word, and the signal that closes the loop. Long-horizon outcomes against a strong baseline — slow, expensive, the only one that counts, the one that feeds the next training pass.

Stated honestly: the causal-AI teacher is itself an approximation, and distillation inherits its biases. We treat the simulator with humility — used for relative ranking of interventions, continuously recalibrated by the closed-loop market tests above. Guardrails are immutable so the actor can't optimise them away.

The product

Three layers, one closed loop.

Stepping back from the model and asking what the product looks like — it's three layers stacked on top of each other, with a closed loop running through all of them. A Data Layer moves brand data in and out. A Cognitive Layer turns it into a continuously-learning behavioural world model. An Agentic Layer is the new UX — opportunity intelligence and autonomous marketing — that lets users drive outcomes through goals and preferences. Each layer consumes from the one below it; the top layer acts on the world; the world's response flows back to the bottom and closes the loop.

THE PRODUCT · THREE STACKED LAYERS · CLOSED LOOP LAYER 03 · AGENTIC The new UX · opportunity intelligence · autonomous marketing Users drive outcomes through goals + preferences; agents propose, simulate and act. Goal & pref UX Opportunity agents Campaign agents Guardrails LAYER 02 · COGNITIVE The behavioural world model · continuous learning Tokenise · encode · sequence · simulate. Profile enrichment lives here. World model Profile enrichment Simulator · causal Continuous learning LAYER 01 · DATA Connectors · inbound & outbound integrations Brand data flows in (web, CRM, edge, market); activation + responses flow back. Inbound connectors Outbound activation Identity & consent Response capture CONSUMES CONSUMES Activation hyper-personalised content User · world receives · responds the real moment of truth Response signal flows back into Data Layer CLOSED LOOP DATA → COGNITION → AGENCY → ACTIVATION → RESPONSE → DATA
01Data Layer

The way data moves in and out

Connectors and integrations for both inbound (brand data flowing in — web logs, CRM, edge, market) and outbound (activation flowing out, plus the responses the world sends back). The plumbing that makes everything else possible.

  • Inbound connectors · web · CRM · edge · market
  • Outbound activation channels
  • Identity, consent & governance
  • Response capture — the loop's return path
02Cognitive Layer

The behavioural world model lives here

This is where the model continuously learns from the data and from the world. Tokenisers, encoders, the seq2seq world model, the simulator — and the profile enrichment that turns raw signal into a usable user-state.

  • Behavioural world model · seq2seq
  • Profile enrichment & latent state
  • Simulator · causal inference
  • Continuous learning from market response
03Agentic Layer

The new UX · outcomes through goals & preferences

The user-facing layer. A new UX moving teams towards opportunity intelligence and autonomous marketing excellence, driven by the user's goals and preferences. Deep agentic enablement — users tell us the outcome; agents propose, simulate and act on it.

  • Goals + preferences UX
  • Opportunity intelligence agents
  • Campaign / activation agents
  • Guardrails & oversight

Data feeds cognition. Cognition feeds agency. Agency acts on the world. The world responds. The response feeds data. One loop, three layers — the product is the shape that loop takes.

Intent AI · From token, to sequence, to world model, to simulator · The vision, for internal review.
The roadmap · how we build it

From the model to the product. Three layers, in a year.

The roadmap builds the three-layer product described in How: a Data Layer for connectors in and out, a Cognitive Layer where the behavioural world model continuously learns, and IntentOne — the product surface where every capability reaches users, through agents and applications. v1 is a telco pilot: prove the closed loop end-to-end with the data a telco already has — web logs, CRM, and time. Every span ships a commercial capability; the loop closes with a controlled-group test.

Data · Cognitive · IntentOne — three layers, one closed loop, a capability at every span.
The shape of the work

Seven workstreams, grouped under the three layers.

Every workstream below maps directly to a piece of the architecture in How. The Data Layer is one workstream. The Cognitive Layer is five — Knowledge Base & Vocab, Encoders, Edge Signal Discovery, Predictor, Simulator. IntentOne — agents and applications — is one. Each is delivered end-to-end (planned · built · evaluated · integrated) and ships a commercial capability when it lands.

Layer 01 · Data

Brand data in, activation out, response back.

Connectors and integrations · the plumbing the rest sits on.
01 · Data Layer

Connectors & integrations

Inbound connectors for web, CRM, edge and market data; outbound activation channels; identity, consent and the response-capture path that closes the loop back into the system.

Inbound · outbound · identity · consent
Layer 02 · Cognitive

The behavioural world model lives here.

Vocab · encoders · sequence · simulator — and the 0-party enrichment that feeds them.
02 · Knowledge Base & Vocab

Behavioural vocabulary

Reference data — brands · topics · interests · intents — plus the behavioural tokeniser that turns raw signal into discrete tokens. The shared vocabulary every downstream model uses.

The reference
03 · Encoders

Per-modality · one shared sequence

Per-modality encoders — Enc_web · Enc_crm · Enc_time first (telco v1) — fuse modalities into one time-ordered behavioural sequence over the shared vocab. v2 widens to market, finance, retail.

v1 telco · v2 sector
04 · Edge Signal Discovery

0-party enrichment across every channel

Intent's SDK mines consented, first-person signal that enriches all four channels of the token — when, where, what, who. Sits across the channels, not as a fifth.

1P + 0P · cross-channel
05 · Predictor

Seq2Seq world model

Self-supervised, next-token, causally masked — time supplies the supervision. Reads the unified sequence the encoders write and emits a prior over what the user does next.

P(next | history, context)
06 · Simulator

Intervention model · actor · simulator

Distilled from a Causal AI teacher (DR-learner · CATE). Takes the prior + a candidate action, emits P(outcome | do(a)). A controlled-group loop closes onto the world model.

f(prior, action) → uplift
Layer 03 · IntentOne

How capability reaches users — agents and applications.

IntentOne is the product. Every capability below is surfaced here — or it can't be used.
07 · IntentOne

Agents and applications

Agents are how users interact with Intent AI — insights, simulations, optimisations: opportunity intelligence agents that surface what to do, campaign agents that compose and run it. The applications they live in, plus the guardrails for oversight, are what make every capability below actually usable. If it isn't surfaced in IntentOne, it can't be used.

Agents · applications · guardrails
Always-on · evaluation

Evaluation runs through every workstream above — behaviour tokeniser, every encoder, reference data (brands · topics · interests · marketing contact responses), the predictor, the simulator. The test is built before the model is trusted.

Early monetisation

Monetisation leads delivery. The first commercial milestone is a paid telco POC signed in September — sold on today's engine and the evaluation proof, ahead of the new capabilities. Each capability then upgrades what's already sold: when the telco encoders land, agents query the encoded space (latent + boolean filters); when the predictor lands, that becomes behavioural-semantic querying; when the simulator lands, every campaign carries a per-user uplift score. Revenue starts early and compounds.

The plan on one page

A year of work, the three layers built one span at a time.

Seven swim lanes spanning the three product layers — Data Layer (connectors), Cognitive Layer (KB & vocab, encoders, edge signal discovery, predictor, simulator) and IntentOne (agents + applications) — run from Jul 2026 through Jun 2027. Each lane is delivered end-to-end: planned, built, evaluated and integrated into the product before the next span starts. Every completed span ships a capability (★ on the chart) — something we can sell. The bottom lane is the commercial path (◆ revenue milestones); the first lands in September, ahead of the build, so monetisation leads delivery and compounds as the stack matures.

How each lane is delivered

Every swim lane below is delivered end-to-end: features are planned, built, evaluated, and integrated into the product. No half-shipped capabilities — when a span finishes, what comes out is sellable. Reading the chart: ★ = a capability shipped ("Cap" — a span finished, evaluated and integrated, something we can sell); ◆ on the bottom lane = a commercial milestone, an actual revenue event.

Layer · 12 months
Jul '26
Aug
Sep
Oct
Nov
Dec
Jan '27
Feb
Mar
Apr
May
Jun
Data Layer Layer 01 · connectors in & out
Connectorsinbound · outbound · identity · consent
Connectors in & out · identity · consent · activation
Foundation · data flowing
Cognitive Layer Layer 02 · the behavioural world model · five workstreams
KB & Vocabbrands · topics · interests · tokeniser
Reference KB + behavioural tokeniser
Cap 1 · reference knowledge
Encodersper-modality → one shared sequence
Enc_web · Enc_time · Enc_crm — telco v1
Cap 2 · behavioural-semantic queries
v2 horizon · Enc_edge → Enc_market → Enc_finance (transactions) → Enc_retail (shopping)
Cap 3 · edge queries
Edge Signal DiscoverySDK · 0-party across all 4 channels
SDK 0-party signal · when · where · what · who
Cap 6 · 0-party enrichment
Predictorseq2seq world model · self-supervised
Sequences keystone → seq2seq world model
Cap 4 · next-action prediction
Simulatorintervention model · Causal AI teacher
Intervention model · controlled-group loop
Cap 5 · best-offer sim
IntentOne Layer 03 · agents + applications · how capability ships to users
IntentOneagents · applications · guardrails
Application shell · early agent access
IntentOne v0 · agents on encoded space
Agents + applications · goals/prefs UX · guardrails
Cap 7 · IntentOne live
Commercial milestones monetisation · revenue events (◆) · leads the build
Commercial pathrevenue milestones
Sep · paid telco POC
Dec · first new-model close
Mar · outcome-based pricing
Jun · agentic + sector pack
Foundation · Brand integrated · data flowing
~ Nov–Dec 2026

The Data Layer is the foundation — not a numbered capability, but the plumbing the seven build on. Inbound connectors, identity & consent, outbound activation, response capture. Once it's live, brand data flows in and out end-to-end.

Cap 1 · Reference knowledge ready
~ Dec 2026

Unified brands · topics · interests · intents + behavioural tokeniser. Agents and product teams query a clean reference base — the shared vocabulary every downstream encoder, predictor and simulator reads from.

Cap 2 · Behavioural-semantic queries
~ Dec 2026

With the telco encoders shipped — Enc_web · Enc_crm · Enc_time producing one unified time-ordered sequence — agents move beyond boolean filters. They can now slice and dice via behavioural semantics over the latent space (“users whose web/CRM/time pattern looks like X”), before the predictor or simulator is even live.

Cap 3 · Semantic querying over edge data
~ May 2027

Once the edge signal encoder is added, the same semantic querying extends to on-device, 0-party signal — agents can find audiences from real-world rhythm, dwell and location context.

Cap 4 · Next-action prediction
~ Apr 2027

With the seq2seq predictor live, the product can predict what a customer is most likely to do next — revealing short-, mid- and long-term intents and moments. Critically, unlike the v1 implementation, this is an optimisation loop we control: dramatically lower inference cost and far easier to evaluate.

Cap 5 · Best-offer simulation
~ Jun 2027

The intervention model — taught by the Causal AI teacher (DR-learner, CATE) — can play out how each user would respond to a marketing campaign and pick the best offer given that user's state and context. Uplift, not just likelihood. The judgement layer of the Cognitive Layer.

Cap 6 · 0-party context enrichment
~ Feb–Apr 2027

Continuous SDK-sourced signals — what they like, where they visited, when something happened, who they are, and what led to a specific event — flow into every channel of the token (when · where · what · who). Critical to the vision: the 0-party layer everything else gets richer on.

Cap 7 · IntentOne live
~ Jun 2027

IntentOne is the product surface where every capability gets used — agents and applications. Agents are how users interact with Intent AI (insights, simulations, optimisations): opportunity intelligence and campaign agents. Plus the applications they live in and the guardrails for oversight. If a capability isn't surfaced through IntentOne, it can't be used.

Commercial path (◆). Monetisation does not wait for the full stack. September: a paid telco POC, sold on today's engine plus the evaluation proof. December: first new-model commercial close on the opportunity / semantic-query SKU (rides the Data Layer + reference knowledge + telco encoders). Q1 2027: outcome-based, uplift-share pricing once the simulator's controlled-group read is credible. Mid-2027: the agentic UX and the first sector pack (finance / retail). Capabilities ship the product; commercial milestones bank the revenue — and they lead the build.

Indicative timeline · current thinking. These dates reflect how we're sequencing the work today. We expect to refine and update them as we move into execution — discoveries in one span shift the next, and evaluation is the gate on every transition.

Workstream · behavioural vocab + encoders (Cognitive Layer)

v1 telco pilot, then sector expansion.

A behavioural token is only as rich as the encoders behind it. v1 is scoped to the data a telco pilot already has: web logs, CRM, and time — the shared temporal axis that anchors every other modality. That's the smallest set that closes the loop end-to-end. The wider vocabulary — edge, market, finance, retail — is held for v2, once v1's controlled-group test has shown real uplift.

v1 · telco pilot

Three encoders. End-to-end loop. One customer.

scope: web logs · CRM · time

The minimum vocabulary that lets us tokenise telco behaviour, train the sequence model, run the simulator and close a controlled-group test on real users.

Live · v1
Enc_web

Brands and topics from search and browse — the highest-frequency consumer–brand signal a telco has.

Web · telco
Build · v1
Enc_time

The shared temporal axis used by every other encoder — clock, day-of-week, dwell, recency, inter-event intervals. Consistent across modalities.

Cross-modality
Build · v1
Enc_crm

Telco profile — tenure, plan, products, value. Who the user already is to the operator.

Profile · telco
v2 · sector expansion

Widen the vocabulary once v1 has shown uplift.

held for next version

Each v2 encoder either deepens the token or opens a new sector. We add them in order of data-flow and consumer–brand interaction frequency — but only after v1 closes the loop.

v2
Enc_edge

On-device signals from Intent's SDK — rhythm, motion, location, app context. The 0-party enrichment layer.

Device · cross-sector
v2
Enc_market

External context — trends, prices, news, the broader pull on a person at the moment.

Market · cross-sector
v2
Enc_finance

Transaction data — what people actually pay for, not just what they browse.

Finance sector
v2
Enc_retail

Shopping and purchase data — baskets, loyalty and store behaviour.

Retail sector

v1 is small on purpose. Web + CRM + time is the smallest vocabulary that lets us prove the whole loop on a telco pilot. Everything else — the edge layer, the sector encoders — is held until v1 has earned the next step.

Workstream · predictor (Cognitive Layer)

From per-event guesses to a seq2seq world model.

Today each prediction is an inference call over a person's brands and interests — accurate enough, but costly and hard to test. The roadmap replaces it with the seq2seq world model from How: per-modality encoders fuse into one time-ordered behavioural sequence over a shared vocabulary, and the predictor learns P(next | history, context) by next-token prediction — self-supervised, causally masked. The human context model is a stack of layers, and the keystone is the sequence.

01Brands & interestsWhat a person touches, resolved to known entities.Live
02SequencesEvents read in order — the keystone that lets every layer above it predict.The keystone
03TraitsThe slow-moving identity the sequence is read against.
04IntentsMicro, macro and long-term — what the pattern is reaching for.
05RoutinesThe rhythms that turn one-off events into expectation.
06NeedsThe need forming underneath the behaviour.
07Moments of needThe prediction a business can act on — each layer exposed as its own output.

Sequences is the pivot. It is the capability not yet built, and the one every layer above depends on — which is why it sits first on the build, ahead of the deeper encoders.

Workstream · simulator (Cognitive Layer)

Score the action, not just the prediction.

Predicting the moment is half the job; knowing whether acting changes anything is the other half. The intervention model from How — the actor / controller / simulator — takes the predictor's prior together with a candidate action a and emits P(outcome | do(a)). It is distilled from a Causal AI teacher (DR-learner · CATE) and continuously recalibrated by a controlled-group loop. Commercial actions enter the sequence as tokens, so the simulator learns not just what happens next, but what would happen if a brand acted now.

Controlled split

~15% control · ~85% treated

A held-out control against a treated group — the only clean read on cause, not correlation. Feeds the closed-loop reinforcement.

Causal AI · teacher

Distillation, not direct training

DR-learner / CATE estimators teach the intervention model what counterfactuals look like, so inference stays cheap while keeping causal grounding.

What it emits

Δ uplift per user

P(y|do(a)) − P(y). Per-person propensity at the moment, plus the ROI of acting. Together, the opportunity worth spending on.

This is the intervention model the How tab puts to work. The Causal AI teacher + control-group loop is how it earns its judgement — turning a predicted moment into a scored, provable opportunity.

Always-on · evaluation

We build the test before we trust the model.

Evaluation isn't a phase at the end — it runs through every workstream above: the behaviour tokeniser, every encoder, the unified sequence, the predictor, the simulator, the SDK signal discovery, and the agentic layer. The same person who builds the piece builds the test that proves it. Reference data — brands, topics, interests, marketing-contact responses — is the line in the sand we measure everything against.

Always-on · runs through every workstream

Two levels, every time

Every metric works on two levels: does the prediction match held-out reality, and does acting on it produce real uplift against a control group. Without a shared yardstick, two approaches can't be compared — like marking two exams without knowing they sat the same paper.

Prediction accuracy

Held-out reality

Next-token and per-layer accuracy on behaviour the model hasn't seen. Baseline first, then measured, incremental gains.

Business uplift

Lift over control

Measured against a hold-out group. A prediction that doesn't move a real number is an opinion, not an asset.

Credibility checks

Does the whole hold?

Single-profile consistency and automated plausibility scoring across thousands of profiles — so the full picture coheres, not just the average.

Sequencing

The order we build in.

Bottom layer first, then up: Data Layer so brand data can flow. Then the Cognitive Layer piece by piece — knowledge base & tokeniser, the telco v1 encoders, the unified sequence, the seq2seq predictor, the intervention model. Then IntentOne — agents and applications — on top. The Edge SDK signal discovery runs alongside the cognitive work. Evaluation runs through every step.

Now
Data Layer + reference

Connectors in & out, identity / consent. Build the KB + behavioural tokeniser so everything else has a vocab to share.

Next
Telco v1 encoders

Enc_web · Enc_crm · Enc_time producing the unified time-ordered sequence. Agents query latent + boolean before the predictor is even live.

Then
Predictor + simulator

The seq2seq world model, then the intervention model distilled from the Causal AI teacher. Closed-loop control-group test on the way.

Top
IntentOne

Agents and applications — opportunity intelligence and campaign agents, the UX they live in, guardrails. Everything below gets used through IntentOne.

Stated honestly: the seq2seq predictor and the intervention model are not yet built — today's production engine runs per-event inference on live telco data. The roadmap is the move from that to the three-layer product in How, earned one proven span at a time. Capability ships at the end of every span.

Intent AI · Data Layer · Cognitive Layer · IntentOne · The roadmap, for internal review.
The 3-month plan · Aug → Oct 2026

Two tracks. Four milestones.

The execution view of the roadmap: two enrichment tracks — Weblogs and SDK — converging on the behavioural-token layer. Milestone 0 lands next week, milestone 1 closes inside this window, milestone 2 starts, milestone 3 is earned later. Every deliverable ships standalone value into IntentOne on the way, and the whole plan is live in Linear under the Intelligence (IntentAI) initiative ↗.

Baseline · quality · tokens · model — in that order, because each one is the evidence for the next.
Two tracks, four milestones

Signal first, then the model — the tracks converge at every milestone.

Internal stakeholders asked for one thing: brands, topics and interests they can trust. So each track earns its way down the page: publish the truth (milestone 0, next week), fix the signal and prove the improvement in the product (milestone 1), encode it (milestone 2), model it (milestone 3). No milestone closes on analysis alone — each one is something a stakeholder can see.

M0Next week
Linear ↗

Baseline established — the data sets everything is measured on

MTNNigeria · weblogs verizonweblogs

Baseline brands, topics & interests published for MTN Nigeria and Verizon — the client’s own data alongside ours, with our honest assessment: “our perception of the baseline” — and the evaluation framework agreed with QA. First task for the incoming QA analyst; the evaluation proof the September POC sells on. These are the reference data sets: every claim after this point is an improvement on these two, or it doesn’t count.

Weblogs
1
Remove the noise Linear ↗

NSFW and machine/adtech activity filtered before enrichment. The fight is the inversion: ~97% precision/recall on non-user activity vs ~15% on user activity today.

2
Improve the quality Linear ↗

Attribution corrected — credit the brand visited, not the ad call underneath. Measured, explainable uplift vs the baseline for MTN Nigeria and Verizon.

SDK
1
Read SDK signal as behavioural context Linear ↗

Today SDK signal feeds nothing into brands, topics & interests. Structure it into the four boxes — when / what / who / where (+ duration) — per time-step, from Alexander’s framing, with the new hire pairing from day one.

2
Derive brands, topics & interests from SDK alone Linear ↗

The Vivo and Fishka proof, at agreed SDK-only thresholds (~70% of full context), with behavioural patterns — “this person is a runner” — where brand affinity isn’t the natural frame.

M1Sep → Oct · closes in window
Linear ↗

Measured improvement — on the same data sets the baseline published

MTNimproved vs M0 verizonimproved vs M0 vivonew · from SDK fishkanew · from SDK

Side-by-side, before and after, on exactly the data sets from milestone 0: improved brands, topics & interests for MTN Nigeria and Verizon, with explainable evidence chains and QA sign-off — plus first-ever coverage for Vivo and Fishka from SDK signal alone — both benefit directly from SDK IntentAI. Not a notebook result: consumable in audience creation in IntentOne through the existing attribute path, no new UI. Does not close on weblogs alone.

Both tracks · converged
Encoding — behavioural tokens Linear ↗

Cleaned signal from both tracks encoded per time-step over the shared vocabulary — the tokeniser the How tab depends on. The unproven leg of the stool (done before / infrastructure / skills): highest research risk, highest moat — which is why it starts only after milestone 1 has shipped standalone value.

M2Starts October
Linear ↗

Behavioural-semantic queries live

Encoding spec covers both sources; semantic search demonstrated on production-scale data; tokens addressable in our app first, then one opt-in customer — slice audiences by behaviour, not just attributes. Closes beyond this window.

Both tracks · converged
Modelling & simulation Linear ↗

The seq2seq world model trained on behavioural tokens, validated predictively against the MTN workstream (pre→post migration, prepay churn, MoMo churn) with client-data-only baselines as the honest comparison.

M3Beyond the window
Linear ↗

The model earns its place

A predictive read vs baselines; a simulation spike; and a documented decision on whether model-discovered segments complement or eventually replace existing micro-segments — earned by evidence, not assumed.

Critical path

M1 has the SDK track on its critical path. If Alexander’s allocation isn’t confirmed, the fallback is M1 closing on weblogs-in-product with SDK following — the degraded case, not the plan. The four decisions that keep this plan whole: M1 scope (brands, topics & interests only, segments & moments deferred to M2+, not dropped), Tobie’s technical leadership of the pipeline handover, Alexander’s allocation, and the two open hires.

The people asks
Allocation

Alexander → SDK

Already started work in the sequencing space; will benefit from adding SDK experience.

Arriving

Intern + PM

A Cambridge Computer Science MSc intern and an IntentAI PM arrive to accelerate the work here.

Open

Two open hires

Two roles open to keep the plan’s dates honest.

Focus

Sangram’s time

IC here, and only here. He makes design decisions for data ingestion, but no IC work is scoped there — which may mean leaning more on Tobie.

Where to follow it: the plan is tracked in Linear — the Intelligence (IntentAI) initiative — with a project per deliverable (linked above), milestones on each, and the PRD attached to the initiative. Dates are indicative: evaluation gates every transition.

Intent AI · The 3-month plan · Aug → Oct 2026 · Milestone 0 next week, milestone 1 in window, milestone 2 started, milestone 3 earned · For internal review.