Ian Anderson co-founded Malachyte, which builds behavior-intelligence technology for ecommerce personalization. After nearly nine years at Spotify working on personalized experiences, Anderson is applying that experience to help retailers understand shoppers’ behavior in real time.
They are most useful for a founder view on real-time ecommerce personalization and recommendation system tradeoffs.
Best for
Retail technology vendors, agencies, or retailers needing technical sessions or explainers on live behavioral personalization.
Not for
Reach campaigns, consumer product promotion, or standard sponsored placements without documented audience and terms.
Key takeaways
Malachyte launch post proposes shift from batch-updated shopper segments to shared real-time behavior signal.
Spotify work involved embeddings, large recommendation systems, and latency constraints.
Posts emphasize privacy, continuously updated understanding, and collaboration with subject experts.
Check first: Confirm they offer sponsorship, deliverables, rates, exclusivity, inquiry route, and audience evidence.
Main ideas
Platform decay is a tooling problem, not a bad-intent problem · Read people individually in the moment, not by bucket · A few behavioral signals can beat a rich profile
Ian Anderson is known here for co-founding Malachyte after nearly nine years at Spotify, where Anderson says they worked on personalized experiences and recommendation systems.
Anderson’s posts connect that experience to a specific ecommerce problem: rules and fixed segments can miss what a shopper wants now.
Malachyte’s stated approach is to use live behavioral signals across search, recommendations, and merchandising.
Follow Anderson’s work for a founder’s view of personalization, product design, and the tradeoffs involved in building recommendation systems—not as independent proof that Malachyte’s product has achieved its stated results.
Anderson’s distinctive contribution in these posts is linking technical experience to a clear product argument.
The Malachyte launch post explains the proposed shift from batch-updated shopper segments to a shared, real-time behavior signal; an earlier post describes Spotify work involving embeddings, large recommendation systems, and latency constraints.
Anderson also emphasizes redeployable tools, collaboration with subject experts, continuously updated understanding, and privacy.
The posts show the argument and priorities, while leaving performance claims to be tested: the Arize:Observe post announces a talk but does not include it, and the stealth hiring post describes a modeling system still being architected.
Anderson’s recommendation of a former colleague is a different, less technical example of their writing, focused on research-mindedness and clear explanation.
Together, these examples make the work useful to readers weighing how personalization systems should be designed, while the supplied material does not establish product outcomes or cover Anderson’s full public body of work.
Ian Anderson co-founded Malachyte, a behavior-intelligence company for ecommerce personalization, after nearly nine years at Spotify working on personalized experiences.
Read more
Across their posts a consistent set of arguments runs: that platform products drifted away from users because the tools behind them are too opaque and too blunt rather than because anyone intended harm; that personalization should read people individually in real time instead of sorting them into batch-updated segments; that a few behavioral signals can reveal intent without needing a rich profile; that data should stay with the customer; and that AI is only useful when combined with human expertise and continuous learning. They build in public, hiring and crediting collaborators while the company was still in stealth.
Platform decay is a tooling problem, not a bad-intent problem
Anderson argues that platforms have steered away from what users want and that this is eroding trust, but places responsibility less on the developers, product managers, and marketing directors who work on them than on the instruments they are given: tools too opaque to target a change, changes too blunt to avoid collateral damage, and scope too narrow to fix the real issue.
Read more
Since the people building the product usually know when it misses the mark, the failure is structural. That diagnosis is the founding argument for Malachyte.
In "Enshittification Is Over!
(If You Want It)" Anderson writes that they sympathize with platform workers, that those workers did not intend the outcome, and that "a decade of experience and a warehouse of GPUs shouldn't be prerequisites for companies to resonate with users."
They then list four commitments for the new company: a complete and redeployable system with low- and no-code components; craft and collaboration over "piecemeal and pugilistic" building; continuous learning about changing people; and privacy by design.
It reads as a manifesto, stating direction rather than results.
Read people individually in the moment, not by bucket
Anderson argues that ecommerce personalization today is mostly rules and segments: match the shopper to a bucket, update it in batches, and hope the fit is close enough.
Read more
Their alternative is a single live behavioral signal, so search, recommendations, and merchandising all draw on the same current understanding of a person rather than on separate batch processes. The reasoning is that cohort membership loses the thing that matters — what this specific person is doing and wanting right now.
Anderson says they and Sidd Motwani started Malachyte because of what they saw digital retail get wrong, and describes the prior work behind it: at Spotify a real-time behavior intelligence layer had to understand 800M+ users "individually, in the moment, not by cohort."
They rebuilt that layer for commerce with integration "measured in days, not quarters."
Elsewhere they describe rearchitecting Spotify's understanding of users and music and then "popping champagne when we tore that down for an even better one" — a willingness to replace a working model when a better one appears.
The move from Spotify to Malachyte is the through-line: after nearly nine years at Spotify, Anderson left to build the same class of system for retail rather than streaming.
Anderson argues you do not need exhaustive data about a person to understand what they want.
Read more
Sparse signals, read against the structure of what they are doing, can indicate intent proactively — the way a search engine infers which page someone wants from a handful of typed words. The priority is inference from limited behavior rather than accumulation of demographic or historical detail.
Quoting the firm Gradient, Anderson compares Google turning "a few typed words and the structure of the web" into an accurate reading of which page someone wanted with Malachyte's aim of using "sparse behavioral signals to proactively understand what someone is actually looking for."
Anderson endorses that framing and thanks Gradient for featuring Malachyte.
The quoted words are Gradient's description of the company, not Anderson's own phrasing; Anderson adds that they have been arguing that "the internet is breaking" and that discovery and digital experiences are not what they used to be.
Anderson argues that the people whose behavior makes a personalization system work should retain ownership of it.
Read more
The company's stated position is to build the machinery that understands behavior while leaving the systems and their uses with the client. This is presented as a design constraint rather than a compliance add-on.
One of the four Malachyte commitments in the founding post is privacy by design, described as: "We don't want your data.
Hell, ideally we don't get any data.
We're building the machinery to understand it and provide quality experiences, but those systems and their uses are meant to be yours."
The same post argues that platforms have eroded user trust, so the privacy stance is tied to that trust argument.
People change, so fixed segments go stale — and AI alone is not enough
Anderson argues that most off-the-shelf services sort users into segments "where they can never escape," while people change and express that change constantly — "not just by how they engage but how they don't."
Read more
A system therefore has to keep learning rather than assign a stable label. A second, related argument is that large language models "lift the floor tremendously" but are incomplete without subject experts, users, and businesses in the loop; the system is meant to be additive, improving as more people use it.
Both points appear together in the founding post as two of the four commitments: dynamic understanding and continuous learning, and craft and collaboration over piecemeal work. Anderson frames the human input as what keeps AI-assisted personalization from being merely automatic.
Anderson treats building a company as a collaborative, visible activity.
Read more
Rather than staying silent until launch, they posted about technical hiring while still in stealth, and publicly credited partners and co-founders as the company took shape. The stated preference is for shared direction-setting over a single architect handing down a finished design.
In a stealth-stage hiring post, Anderson said they were looking for an engineering partner to architect the modeling system, noted that an initial design was already being built on, and added that they would want that person to craft the roadmap for later stages.
In the Malachyte launch post they name co-founders Sidd Motwani and Shivaditya Sinha, crediting Sinha with scaling early operations.
A separate post recommends a former Spotify intern, Sam, as "thoughtful, multifaceted, research minded," someone who "grows the folk around him" and can explain deep subjects to anyone — evidence of what Anderson values in collaborators.
Answers why Ian Anderson thinks platform personalization broke and what Malachyte says it will do differently.
Read more
Start here for the founding argument.
Anderson describes platforms drifting away from what users want and argues the problem is opaque, blunt tools rather than bad intent.
The post points to their Spotify work on personalized experiences, billion-parameter models, and millisecond latency constraints, then lists four Malachyte commitments: a complete and redeployable system, craft and collaboration over piecemeal work, dynamic understanding and continuous learning, and privacy by design.
It is a manifesto-style post, so it states direction more than proof.
For brands planning a campaign
For a retail personalization brief, these become testable vendor claims: ask how the system stays additive, how it updates user understanding without fixed segments, and what privacy by design means in data handling. Useful for setting evaluation criteria, not for evidence that the product already performs.
Answers how Malachyte connects Spotify-scale behavior intelligence to ecommerce and what it claims to replace.
Read more
Start here for the product pitch.
Anderson says they and Sidd Motwani started Malachyte after seeing digital retail rely mostly on rules and segments: match a shopper to a bucket, update in batches, hope it is close enough.
They describe rebuilding a Spotify behavior intelligence layer that had to understand 800M+ users individually in the moment, now applied to commerce so search, recommendations, and merchandising use the same live signal.
They also credit co-founder Shivaditya Sinha with scaling early operations.
For brands planning a campaign
This directly addresses a real-time ecommerce personalization brief. Use it as a starting hypothesis for what a behavior layer should cover, then ask for proof on integration measured in days, not quarters, and on how one live signal serves search, recommendations, and merchandising.
Answers what Anderson did before Malachyte and why the career change matters.
Read more
Start here for background.
Anderson says they started at Spotify about nine years earlier and decided to leave two months before the post.
The post is reflective rather than technical: gratitude for Spotify, a longer farewell email that exceeded LinkedIn's limit, and appreciation for what passionate teams can do together.
It says more posts about the new thing would follow and asks former colleagues to drop Wrapped soon.
For brands planning a campaign
Establishes depth of platform experience and the timing of the move into a new company. It is not a method source, but it helps assess whether the operator has lived inside large-scale personalization and can speak credibly about product culture.
Answers where Anderson has explained recommendation and embedding pitfalls.
Read more
Start here for technical speaking evidence.
Anderson announces a talk at Arize:Observe and says that during their Spotify time, embeddings grew to power nearly all recommendations.
They describe the talk as covering challenges and pitfalls that arise when delivering audio content for half a billion people every month, and they expect embeddings to become more influential as AI progresses.
The post is an announcement, not the talk itself.
For brands planning a campaign
Evidence of public technical communication on embeddings and recommendation systems.
Useful for judging whether the founder can explain tradeoffs to technical buyers.
Ask for the talk recording or slides because the post alone does not contain the challenges or fixes.
Answers what technical role and modeling system Malachyte was building at that point.
Read more
Start here for team-stage context.
Anderson says they are hiring while still in stealth and are looking for an engineering partner to architect the modeling system.
The post notes an initial design is already being built on, but Anderson wants collaboration and wants that person to craft the roadmap for later stages.
It points to a job description for more detail.
For brands planning a campaign
Shows the modeling system is early and actively being architected. Useful for technical diligence and timeline expectations, but it is not evidence of a finished platform or a proven production system.
Answers what Anderson values in researchers and colleagues.
Read more
Start here for working-style evidence.
Anderson says they first met Sam when Sam joined as an intern in 2016, and describes Sam as thoughtful, multifaceted, and research minded.
The post credits Sam with growing the people around him, explaining deep subjects to anyone, and finding ways to enlighten in spaces others thought were established.
Anderson closes with a direct recommendation to hire him.
For brands planning a campaign
Not a campaign-fit source, but evidence of how Anderson evaluates research talent and collaboration. It can inform whether a partner values research rigor and mentorship on a personalization engagement.
Answers how Malachyte describes its core idea to outsiders and who has featured it.
Read more
Start here for positioning and outside reception.
Anderson quotes a Gradient piece comparing Google's use of typed words and web structure with Malachyte's use of sparse behavioral signals to understand what someone is looking for.
Anderson thanks Denise T. and Gradient for featuring Malachyte and says they have been arguing for years that the internet is breaking and that discovery and digital experiences are suffering.
The post links to the Gradient blog but does not include its full text.
For brands planning a campaign
Third-party framing of Malachyte's core analogy. Useful for understanding external positioning, but the actual Gradient article was not supplied here, so the quote should not be treated as independent performance evidence.
Platform trust, opaque tools, and privacy-by-design1
Sparse behavior signals for discovery1
People development and endorsement voice1
Real-time behavior intelligence for commerce
Anderson argues that ecommerce personalization is still mostly rules and segments: shoppers are matched to buckets and updated in batches.
Read more
They say they had already built a real-time behavior intelligence layer at Spotify that had to understand 800M+ users individually in the moment, and that Malachyte rebuilt that layer for commerce as one system powering search, recommendations, and merchandising from the same live signal, with integration measured in days rather than quarters.
Unified live behavioral signal powering search, recommendations, and merchandising.
For brands planning a campaign
A retail technology company can assess whether Anderson's stated differentiation—a unified live behavioral signal rather than batch segmentation—maps to its real-time personalization gap. The strongest fit is a brief focused on replacing or augmenting rules-and-segments personalization with one system across search, recommendations, and merchandising.
In a talk description, Anderson says that during their time at Spotify, embeddings grew to power nearly all recommendations and that delivering audio content for half a billion people every month creates real challenges and pitfalls.
Read more
Another post says they optimized billion-parameter models and millisecond latency constraints.
Embeddings at scale, large models, and low-latency constraints.
For brands planning a campaign
Technical evaluators can probe Anderson's depth in embeddings, latency, and recommendation failure modes. This is useful where a retailer needs to scale personalization without losing speed or relevance, especially if the internal team will ask detailed architecture questions.
Platform trust, opaque tools, and privacy-by-design
Anderson argues that platforms have steered away from what users want and that this is eroding trust.
Read more
They place responsibility less on individual developers or product managers than on tools that are too opaque, changes that are too blunt, and scope that is too narrow. Malachyte's stated approach includes modular, low- and no-code components; LLM assistance combined with expert and user input; continuous learning about changing people; and being private by design, with the goal of not wanting customer data.
The posts distinguish no documented offer of sponsorship from claims about system performance; they describe intended principles and early shipping, not audited results.
For brands planning a campaign
A brand can evaluate a trust-and-privacy positioning for personalization. If a retailer is worried about data governance or customer trust, Anderson's arguments and stated system principles give useful talking points, but the posts also frame these as Malachyte's intended direction rather than proven outcomes at scale.
Quoting a Gradient feature, Anderson compares Google's ability to turn a few typed words and web structure into accurate page understanding with Malachyte's aim: use sparse behavioral signals to proactively understand what someone is looking for.
Read more
Anderson also says they have been arguing that the internet is breaking and that discovery and digital experiences are not what they used to be.
For brands planning a campaign
This gives a concrete frame for retailers trying to improve search, discovery, and merchandising from limited behavioral data rather than only from rich profiles or historical segments. A campaign brief can test whether that sparse-signal approach fits the retailer's data constraints.
While still in stealth, Anderson posted openly about hiring an engineering partner to architect the modeling system.
Read more
They said they had an initial design already being built on but wanted collaboration on the roadmap for later stages. Anderson also publicly thanked Gradient for featuring Malachyte and named co-founders Sidd Motwani and Shivaditya Sinha, with Sinha credited for scaling early operations.
For brands planning a campaign
The posts show an early-stage, collaborative, public-building posture.
A brand considering co-development, a pilot, or a founder-led partnership can weigh that against the need for mature vendor proof.
The public hiring and co-founder details also give specific signals about team structure and current priorities.
Anderson wrote a recommendation for Sam, who joined as an intern in 2016, describing Sam as thoughtful, multifaceted, research-minded, able to explain deep subjects to anyone, and able to grow the people around him.
Read more
This post concerns a colleague, not Anderson's own personalization work or commercial offer.
For brands planning a campaign
This is not a product claim, but it shows a collaborative talent-development voice that can matter when a brand evaluates whether Anderson is a credible technical partner or ambassador. It is weaker evidence for direct campaign fit than the Malachyte and personalization posts.
Ian Anderson's documented business relationships centre on Malachyte, the behaviour-intelligence company he says he co-founded for ecommerce personalization, and on Spotify, where his own posts place nearly nine years of work on personalization and recommendations.
Read more
The remaining connections in the reviewed posts are mentions: a conference talk at Arize:Observe and Gradient featuring Malachyte in a blog post. None of the posts describe sponsorship, affiliate, or client arrangements.
Malachyte
Owned business
Anderson states that he and Sidd Motwani started Malachyte, which builds a behaviour-intelligence layer for ecommerce, and describes the system as one he and his co-founders are shipping.
Read more
He names Shivaditya Sinha as a co-founder running early operations. The posts describe product scope — search, recommendations, merchandising off a shared live signal — rather than a client list.
Anderson's own posts say he started at Spotify about nine years before writing, worked on personalized experiences and embedding-based recommendations, and left two months prior to that farewell post to start something new.
Read more
Earlier posts describe building new personalized experiences and researching their impact, and delivering a talk on embeddings for audio content reaching half a billion monthly listeners. The posts establish the employment and the work; they do not carry dates.
Anderson says he delivered a talk at the Arize:Observe event, covering the challenges of embeddings when serving audio content at half a billion monthly listeners.
Read more
The post documents the speaking engagement at the company's event and no other commercial arrangement.
Ian Anderson's posts read like a builder's public notebook.
Read more
Start with the Malachyte launch post for the central argument and product direction, then go to the Spotify-era posts for the behavior-intelligence and embeddings background behind it.
The stealth hiring and collaborator posts show how Anderson builds teams and shares credit.
The through-line is personalization that reads people individually in real time, uses sparse behavioral signals instead of batch segments, and keeps customer data with the customer.
Treat the posts as early-stage company framing and technical advocacy, not independent product audits or long-run results.
Cohn's posts also focus on retail technology, including AI for personalized product recommendations and tailored customer journeys.
Read more
He describes a Shopify migration through unified customer data, flexible fulfillment, and conversion gains, which overlaps with Anderson's interest in ecommerce personalization and live customer understanding.
Cohn writes from inside Shopify as an implementation partner, using named retailer outcomes and platform case studies.
Anderson writes as a startup founder arguing for a new behavior-intelligence layer and questioning platform incentives.
Cohn is more case-study oriented; Anderson is more architecture and argument oriented.
Moock's posts follow AI-driven commerce infrastructure, including making catalogs discoverable to AI agents and how distribution complexity affects merchants.
Read more
That shares Anderson's interest in how shoppers discover products and how commerce systems need new infrastructure for intent.
Moock focuses on agentic commerce, checkout, and payment and catalog plumbing at Stripe.
Anderson focuses on behavior intelligence for personalization across search, recommendations, and merchandising.
Moock's frame is channel readiness and integration; Anderson's is understanding the individual shopper in real time.
McIntosh writes about bringing AI into retail stores, connecting online and in-store data into a personalization flywheel, and using shelf and shopper signals to understand choices.
Read more
That is close to Anderson's argument that commerce should read behavior signals to understand what someone wants.
McIntosh's evidence is physical retail AI: computer vision, shelf intelligence, smart carts, and Instacart's acquisition of Arpalus.
Anderson's is a software behavior-intelligence layer for digital commerce.
McIntosh is operational and in-store; Anderson is personalization infrastructure and startup building.
For a retail-personalization brief, Anderson’s posts suggest relevant subject matter: they discuss applying behavior intelligence to ecommerce search, recommendations, and merchandising, and describe moving from Spotify personalization work to Malachyte.
Read more
That shows what Anderson talks about, not who follows or engages with the content.
The supplied evidence contains no dated audience-composition data or platform performance metrics.
Before choosing a placement, request recent audience results and, if the brief requires professional targeting, a documented survey or other evidence of followers’ roles, industries, and seniority.
The reviewed material contains no documented partnership or sponsorship program for Ian Anderson.
Read more
Seven LinkedIn posts cover Malachyte's launch, its behavior-intelligence product for ecommerce personalization, hiring for an engineering role, a talk on embeddings at Spotify scale, commentary on platform incentives and internet discovery, and a public recommendation of a former colleague.
None of them announce a sponsored placement, a brand collaboration, a media kit, rate information, or a way to submit a partnership inquiry.
That absence is a research gap rather than a statement that Anderson never accepts sponsorship work.
Practically, a brand planning a campaign has nothing here to budget against: no disclosed format, deliverable, audience specification, exclusivity term, or turnaround commitment.
Any outreach would have to start by asking directly what, if anything, Anderson offers commercially, and the answer would need to be confirmed before a plan is built around it.
Strong fit for a brief that needs technical credibility on real-time personalization, not mass reach.
Read more
The clearest case is a retail-technology vendor, agency, or retailer running a technical session, co-authored explainer, or pilot-level advisory engagement about replacing batch segmentation with a live behavioral signal.
Anderson is a direct subject-matter match: the posts describe rebuilding a Spotify behavior intelligence layer that had to understand more than 800 million users individually and in the moment, then applying that layer to commerce so search, recommendations, and merchandising run off the same live signal, with integration measured in days rather than quarters.
They also argue publicly about why platforms drift from user intent and what they say a better system should do, including privacy by design and continuous learning rather than fixed segments.
The Arize:Observe talk announcement and the mention of billion-parameter models and millisecond latency constraints show they are willing to explain dense recommendation and embedding tradeoffs in public, which is the skill a technical personalization brief actually needs.
Two conditions belong in the plan.
First, the reviewed material documents no sponsorship format, deliverable list, rate card, exclusivity term, or inquiry route, so terms have to be established directly before any budget is committed.
Second, the audience evidence is thin: the account listing shows roughly 1,005 followers and no composition data, so this brief should be scoped around expertise delivered through a shared session or written collaboration rather than around reach.
When to choose a specialist
Weak fit, and likely a decline, for a brief that buys audience reach, consumer-style product promotion, or a standard sponsored placement.
Read more
The reviewed posts show no sponsored example, no media kit, no format or rate information, and no way to submit a partnership inquiry, so there is nothing here to budget against.
Reach is small: the account listing shows about 1,005 followers, with no dated demographic or engagement evidence, so a brand cannot show that a placement would land with shoppers or even with a defined professional segment.
The public voice is also wrong for that brief: the posts are founder-manifesto and technical, not consumer reviews or lifestyle content, which makes them a poor vehicle for retail product awareness.
There is a conflict dimension too.
Anderson runs a company selling ecommerce personalization, so a brand that competes in or buys that category would be asking a vendor in the space to promote its product, with disclosure and competitive-positioning problems that a standard influencer contract does not resolve.
One more practical limit: only two of the seven posts carry dates, so posting cadence and current activity cannot be established from this material.
If a buyer still wants to pursue a reach-based brief, the honest starting point is a direct question about whether any commercial format exists and what it would include, plus fresh audience data.
A worked campaign example
Brief: a retail-technology company wants to improve real-time personalization on its ecommerce platform, and wants Anderson's involvement in shaping the effort.
Worked angle: a technical working session and a short co-authored briefing built around one testable comparison, batch segments versus a live behavioral signal.
Shape it like this.
Open with Anderson's own framing from the founding post: personalization is still mostly matching a shopper to a bucket, updating in batches, and hoping it is close enough, while their claimed alternative is one behavior layer feeding search, recommendations, and merchandising from the same live signal.
Then hand the retailer's team a structured half-day agenda that starts from its actual constraints: which behavioral events it already captures, how often its segments refresh, where search, recommendations, and merchandising currently make separate decisions, and what latency its pages can tolerate.
Use Anderson's stated principles as the evaluation criteria rather than as proof: whether the system stays additive as it is used, whether it updates an individual's understanding without locking them into a segment, how sparse behavioral signals can stand in for rich profiles, and what privacy by design means for data handling and retention in practice.
Close with a shared one-page checklist the retailer can run against its current stack, plus a named next step such as a two-week scoping exercise.
Set expectations honestly in the pitch: the posts describe direction and early shipping, not audited performance, the Gradient article and the full Arize:Observe talk were not supplied so those claims need source material before they are repeated, and the brand should confirm commercial terms directly because no offer, rate, or deliverable is documented.
This angle uses the exact evidence that makes Anderson relevant without promising results or reach the evidence does not support.
Choose Cohn when the brief is about putting personalization to work on a commerce platform and the team wants an operator's account of what changes inside a retailer.
Read more
The post on PAIGE's move to Shopify walks through the practical effects of unifying customer data across channels: stores fulfilling 17% of online orders in a peak season, retail staff training cut from two weeks to one or two days, and a 50% lift in online conversion.
A follow-up article co-authored with DEPT covers using AI for personalized recommendations and tailored journeys.
That is the right evidence when the decision is platform selection, merchandising process, and an adoption plan the retailer can start this quarter.
Cohn explains outcomes and vendor partnerships, not the machinery behind them.
A team that needs to design a live behavioral signal layer, weigh embedding approaches, or reason about latency will find no model-level detail here; Anderson's talk announcement on embeddings at Spotify scale and the Malachyte behavior-layer framing sit closer to that work.
Cohn's posts also read as platform-side advocacy, so the personalization case arrives through a migration story rather than a vendor-neutral test.
Treat Cohn as the operator's view of the same decision and Anderson as the engineering view, not as substitutes.
Choose Moock when the retailer's real problem is discovery and distribution: getting its catalog found and bought through AI assistants and marketplaces rather than inside its own storefront.
Read more
The Stripe posts describe a single integration that makes an entire catalog discoverable to a range of AI agents, with checkout, payment, and fraud handling built in, and a measured reading of OpenAI moving Instant Checkout from ChatGPT search into Apps: people used AI to research products but did not buy through chat, discovery works, and the durable win is open protocols, catalog syndication, and shared payment tokens across agents. That is the sharper brief when the campaign question is how to become agent-ready before the channel matures.
Moock sits one layer away from in-session personalization.
His evidence covers catalogs, checkout plumbing, and agent channels, not understanding an individual shopper in the moment, so he cannot answer the part of the brief about which live signals change a recommendation.
He also speaks from a payments platform, and the posts are launch- and announcement-adjacent, so a buyer gets an infrastructure roadmap rather than a tested personalization exercise.
If the goal is preparing catalogs for agent traffic, Moock is the better first call; if it is running a behavioral-signal test on your own ecommerce experience, Anderson is closer to the work.
Choose McIntosh when personalization has to reach the physical store and the company runs in-store or omnichannel operations.
Read more
The Instacart posts cover the Arpalus acquisition and its shelf-recognition technology working on-device in real store conditions, combined with a shopper network that visits stores daily, feeding a data flywheel into fulfillment, an AI assistant, and Caper.
The Forbes post makes the argument directly: winners will use Physical AI to learn where customers pause, what they substitute, and what convinces them to pick one item over another, from smart carts and shelf intelligence to fixing out-of-stocks.
That is the right evidence when the brief's real blind spot is shelf data rather than the digital session.
McIntosh's material is grocery and in-store physical AI, and it arrives largely as corporate acquisition and media announcements rather than a reusable personalization method.
It also assumes stores and a shopper fleet to instrument; an ecommerce platform with no physical footprint cannot use that layer, and for the online session he offers less than Anderson.
The tradeoff cuts both ways: Anderson's posts stay on the digital behavior layer and would not answer a brief whose gap is in-store data, so pick McIntosh when the shelf is the problem and Anderson when the live online signal is.
This profile draws on seven LinkedIn posts written by Ian Anderson, read in full, and the accepted overview, ideas, relationships, discovery, and fit sections.
The posts cover Malachyte’s launch and product direction, a platform-incentives argument, a Spotify departure reflection, an Arize:Observe talk announcement on embeddings, a stealth-stage hiring post, a colleague recommendation, and a Gradient feature.
Two posts carry dates, July 2025 and August 2026; five were supplied without dates.
Not supplied: the full Arize:Observe talk, the Gradient article, the linked job description, audience demographics or engagement, and any rate, format, exclusivity, or inquiry evidence.
Those remain gaps for a buyer to close before committing to paid work.
Recent engagement
Across their last 7 LinkedIn posts, the median post got 88 reactions and 3 comments. Checked 2026-11-02.
