David Arnoux

His alternative starts with three prerequisites — executive buy-in with urgency and a budget, the right team, and aggressive prioritisation that says no to everything else for now.

• One AI centre of excellence owns guardrails, governance, cross-unit trade-offs, leadership education, and shared components, meeting monthly.
• Each business unit gets a small embedded team: an executive sponsor, an internal product owner, someone close to the tooling, and an AI-native engineer, meeting weekly.
• Use cases sit on one ranked backlog scored on impact and ease.
• He argues about 80% of the stack — data layer, connectors and access, agent runtime, guardrails and audit, patterns and evals — should be built once and reused, with the last 20% fitted per use case.

In the post, scored use cases move through scope, prioritise, proof of concept, controlled pilot, pilot, harden, monitor and optimise. The centre of excellence owns the shared data layer, connectors, agent runtime, guardrails, audit, patterns and evals, and runs on whatever cloud the company already uses.

Read original

Proprietary logic points to building.

• Team-wide reliability points to hiring an off-the-shelf agent.
• In both examples he keeps a human decision on outbound actions, with scoped permissions and an audit trail.

A revenue operations team hires an agent into Slack that reads every open deal overnight, posts the pipeline review before Monday standup, flags forecast gaps the day they appear instead of at month end, and answers executive questions in about 90 seconds with the source attached. A five-person team gives another agent partner outreach: it reads replies from two tools at 9am, sorts them by intent, drafts follow-ups in the team's voice, and posts one summary, with nothing sent until a teammate approves it in the thread.

His conference notes later call constant human-in-the-loop review largely a myth once guardrails, compliance, security, and branding are designed in, while these agent deployments still keep human sign-off before outbound actions.

Read original · Original 2

An agent reads price, coverage and quality and picks, so a product whose price sits behind a demo call gets skipped. He says agents need data without contracts or minimum spends, and that pay-as-you-go replaces subscriptions.

His product routes enrichment calls across more than 100 providers on quality and cost, charges per use, and bills only when a call returns a result.

• One monthly update reports 460 sign-ups, 247 paying customers, over a million calls, 95 providers used, customers in more than a dozen countries, and teams cutting data bills 30 to 40%.
• Partner posts add website technology profiles at about 5 cents a call and verified work emails at about 4 cents, with no charge on a miss.

Read original · Original 2 · Original 3 · Original 4

Fast and strong are network and intros, executive dinners, warm LinkedIn DMs, signal-based outbound and webinars.

• Slow and strong are founder posting, SEO, AI search, partnerships, community and free tools, which take six to twelve months before paying and then compound.
• Fast and weak are paid search, LinkedIn ads, creators, cold calling, cold DMs, Product Hunt and AI SDRs, suited to capped bursts.
• Slow and weak, which he says to skip, are PR, big conferences, owned newsletters and generic cold email.
• He reasons that AI has cut the cost of running the strong motions, so a small team can keep the top half running at once.

He reports findings from a survey of more than 100 GTM experts plus his own GTM OS data checked against 2025-2026 benchmarks.

• The sharpest point: generic cold email and signal-based outbound share the same mailboxes and the same copy, but the buying signal is the whole difference.
• He also notes podcasts were barely mentioned and that PR brings little pipeline on its own while feeding SEO and AI search citations.

Read original

His argument is that everything above the foundation wraps the layers below, that the chokepoints — a handful of firms in a few countries — carry the fattest margins, and that the biggest bottlenecks sit at GPUs and data centres. He says a buyer of AI is buying the top of the stack, while price changes, outages and rate limits often start lower down.

He names the most advanced lithography machines and the largest foundry as chokepoints, cites big tech guiding toward $725B of 2026 capex mostly for AI, and notes a new data centre still takes years to build.

• He says he would put spare cash into GPUs and data centres.
• The same constraint argument appears in his reading of Anthropic's S-1 figures: more than 11,900 customers now spend over $100K a year, up from about 950 a year earlier, and the binding limit is compute, power and data centres rather than demand.

Read original · Original 2

He argues the human-in-the-loop idea is largely a myth once guardrails, compliance, security and branding are designed in, because the volume of activity is too high for people to review at every step. He also stresses that distribution still decides outcomes when everyone can build.

He writes that prototype-to-fully-deployed work can be done in two weeks even for enterprises, and that traditional companies are building software from scratch for their existing customers instead of buying it. His proposed alternative to constant review is strong guardrails, compliance, security and branding by design.

The position sits in tension with his own agent case studies, which still keep human approval before sending outbound replies or actions.

Read original

This post lays out his operating model, from executive sponsorship and a shared center of excellence to weekly prioritization and reusable infrastructure.

The post contrasts two approaches Arnoux says fail: centrally managed programs that produce slides but do not ship, and isolated team experiments that do not reach production or get reused.

• His alternative assigns a small team to each business unit, ranks use cases by impact and ease, and moves them through defined stages from scope to monitoring.
• He proposes building shared data access, agent runtime, guardrails, and evaluation patterns once, then adapting them to each use case.
• This is a concrete guide to the organizational work around AI, not just a list of tools.

For brands planning a campaign

For a campaign about enterprise AI implementation, use this as evidence of the issues Arnoux emphasizes: ownership, prioritization, governance, and reuse. Ask him to show how a team would choose and advance its first use case.

This post gives a specific example of an agent handling pipeline review and forecast questions, including the controls described for the deployment.

The post starts with a recognizable RevOps problem: deal changes are spread across tools, so forecast issues surface late.

• Arnoux describes a client using an off-the-shelf agent to review open deals overnight, flag forecast gaps, and answer questions in Slack with a source attached.
• He frames the build-versus-buy choice around whether the job depends on proprietary logic or reliable operation across a team's tools.
• The example also specifies scoped permissions, audit trails, and human approval before outbound actions.
• The post is explicitly marked sponsored, so it is both a workflow example and a product promotion.

For brands planning a campaign

A RevOps or agent-platform campaign could ask Arnoux to walk through the workflow, the approval boundary, and how a buyer should test reliability across tools. Keep the sponsorship context clear and distinguish the post's reported outcomes from independently verified results.

This post groups go-to-market channels by speed to payoff and potential pipeline, making his priorities and tradeoffs easy to inspect.

Arnoux sorts channels into four groups: fast and strong, slow and strong, fast and weaker, and slow and weaker.

• He puts warm introductions and signal-based outreach in the first group, while founder posting, SEO, partnerships, community, and free tools are longer-term bets.
• He cautions that the map is aggregate and that market, deal size, and company stage can shift the results.
• The useful distinction is between channels that can produce quickly and those that compound over time, rather than treating every channel as interchangeable.

For brands planning a campaign

For a B2B growth or marketing-technology campaign, this is a starting point for asking how he would adapt channel choices to a specific market and sales cycle. The post's channel rankings are presented as aggregate findings, not a prescription for every company.

This post walks through the infrastructure, compute, and software layers Arnoux says matter.

Arnoux moves from energy, materials, chip manufacturing, and data centers up through clouds, models, and applications.

• He uses the layered structure to explain how supply constraints lower in the stack can affect products higher up.
• He then connects the model to a buyer's concern: price changes, outages, or limits in an AI product may originate in its underlying dependencies.
• The post says an accompanying visual illustrates the stack; the available text describes it, but does not establish what the visual itself shows.

For brands planning a campaign

For a cloud, infrastructure, or AI procurement campaign, use this piece to frame a discussion about dependencies and bottlenecks beneath the application layer. Avoid treating its investment opinions or market claims as independently established evidence.

This post describes how partnership replies are sorted and drafted before the team starts its day.

Arnoux describes an agent reading replies from two outreach tools, sorting them by intent, drafting responses in the team's voice, and posting a summary to Slack.

• He says teammates can request pipeline information there, but that a person must approve a reply before it is sent.
• The example makes the division of labor clear: the agent prepares and organizes the work; a team member retains the sending decision.
• The post promotes a named tool and is marked “Paid Partnership.”

For brands planning a campaign

For a small-team automation campaign, this is a useful prompt for discussing where an agent can save time and where approval remains important. Preserve the paid-partnership disclosure and treat the stated workload figures as the post's own account.

It starts with three prerequisites: executive buy-in with real urgency and a budget, the right team, and aggressive prioritisation that says no to everything else for now.

• One AI centre of excellence owns guardrails, governance, cross-unit trade-offs, leadership education, and shared components, and meets monthly.
• Each business unit gets a small embedded team with an executive sponsor, an internal product owner, someone close to the tooling, and an AI-native engineer, meeting weekly.
• Use cases sit on one ranked backlog scored on impact and ease and move through scope, prioritise, proof of concept, controlled pilot, pilot, harden, and monitor.
• Roughly 80% of the stack — data layer, connectors and access, agent runtime, guardrails and audit, patterns and evals — is built once and reused, with the last 20% fitted per use case.
• He names two paths that fail: a central top-down programme that ships nothing, and scattered per-team proofs of concept that never reach production.

Five-part model presented with a diagram and contrasted against two common failure paths.

For brands planning a campaign

A platform, enablement, or transformation vendor can sponsor a deeper walkthrough or workshop on the centre-of-excellence model. The build-once-reuse story favours infrastructure and platform positioning over single-point tools.

Examples: Source 1

Fast and strong "sprints": network and intros, executive dinners, warm LinkedIn DMs, signal-based outbound, webinars.

• Slow and strong "marathons": founder posting, SEO, AI search, partnerships, community, free tools — six to twelve months before they pay, then compounding.
• Fast and weak "HIITs": paid search, LinkedIn ads, creators, cold calling, cold DMs, Product Hunt, AI SDRs, suited to capped bursts.
• Slow and weak to skip: PR, big conferences, owned newsletters, generic cold email.
• He argues AI has cut the cost of running the strong motions, so a small team can keep the top half running at once, and that generic cold email and signal-based outbound share the same copy but differ entirely by buying signal.

Two-by-two speed-versus-pipeline matrix, explicitly framed as aggregate that shifts with market, deal size, and stage.

He notes podcasts were barely mentioned by respondents, and that PR, while slow on its own, feeds SEO and AI search citations.

For brands planning a campaign

Useful for content that helps buyers allocate channel spend, or for a sponsored deep dive or webinar on one motion. The matrix format also supports interactive budget-allocation tools and lead magnets.

Examples: Source 1

Hidden pricing behind demo calls and "contact sales" gets skipped by an agent, so prices and coverage need to be machine-readable.

• His own product routes enrichment calls across more than 100 providers on quality and cost, charges per use, and bills only when a call returns a result.
• One month after launch he reports 460 sign-ups, 247 paying customers, over a million calls, 95 providers used, customers across more than a dozen countries, and teams cutting data bills 30-40%.
• Partner posts add website technology profiles at about 5 cents a call and verified work emails at about 4 cents, with no charge on a miss and about 39 cents to add a mobile number.

Launch metrics posts plus provider partnership announcements, each describing what a specific call returns and what it costs.

For brands planning a campaign

Data providers, APIs, and infrastructure vendors can evaluate him for launch narratives, partner announcements, and comparison content aimed at agent builders. His metric posts show a template for turning usage data into a credible growth story.

Examples: Source 1 · Source 2 · Source 3 · Source 4

Two posts marked as paid partnerships illustrate the buy side.

• In one, a revenue operations team hires an off-the-shelf agent into Slack that reads open deals overnight, posts the pipeline review before Monday standup, flags forecast gaps the day they appear instead of at month end, and answers executive drill-downs in about 90 seconds with the source attached.
• Scoped permissions, human sign-off on every outbound action, and an audit trail are what got it past internal compliance review.
• In the other, a five-person team gives the agent partner outreach: reading replies across two tools at 9am, sorting them by intent, drafting follow-ups in the team's voice, and posting one summary, with nothing sent until a human approves in the thread.

Decision rule of thumb plus first-person adoption narrative; both posts disclose a paid partnership.

Both examples are promotional placements, so the recommendations are made in that context rather than as neutral comparisons.

For brands planning a campaign

Agent, RevOps, and partnership tool vendors can evaluate him for adoption case studies and sponsored posts that show an agent taking one specific, measurable job. The build-versus-buy framing gives a sponsor a clear slot as the buy option.

Examples: Source 1 · Source 2

His argument is that everything above the foundation is a wrapper on the layers below, that the chokepoints — a handful of firms in a few countries, such as the most advanced lithography and the largest foundry — carry the fattest margins, and that the biggest bottlenecks sit at GPUs and data centres. He cites big tech guiding toward $725B of 2026 capex, mostly for AI, notes a new data centre still takes years to build, and says he would put spare cash into the GPU and data-centre layers.

Layered visual diagram built around a mining metaphor and animated with an AI coding assistant.

For brands planning a campaign

Infrastructure, cloud, and compute vendors can evaluate him for explainer content that places their product in the stack. The chokepoint framing also supports sponsored market analysis on capex, power, and data-centre constraints.

Examples: Source 1

He notes more than 11,900 customers now spend over $100K a year, up from about 950 a year earlier, and argues the binding constraint is compute, power, and data centres rather than demand. He highlights that no traditional venture firms were in the Series A or B, and that FTX's roughly $500M Series B stake was sold by the estate for about $1.3B in 2024, then asks whether the company will absorb the knowledge economy.

Cube visual comparing the Series A against the implied outcome, built from S-1 figures.

For brands planning a campaign

Financial, venture, and market-data brands can evaluate him for commentary that turns filings into visual, shareable analysis. The format suits data-visualisation partnerships.

Examples: Source 1

He argues the human-in-the-loop idea is largely a myth once guardrails, compliance, security, and branding are designed in, because the volume of activity is too high to review at every step, and he stresses that distribution still decides outcomes in a build-everything world.

Conference field notes in bullet form, mixing observations with links to his own products and takeaways.

For brands planning a campaign

Event sponsors and developer-tool brands can evaluate him for conference recaps and adoption-curve commentary. His position against constant human review is a sharp, quotable stance for campaigns targeting operations leaders.

Examples: Source 1

He reports month-one figures — 247 paying customers, 460 sign-ups, over a million calls, five-figure revenue — and closes the anniversary post with thanks "from our team (Walid Boulanouar, Luísa Lima, Adel Dahani and me)."

• Later posts speak as "our five-person team at looot."
• His stated thesis behind it: pay-as-you-go replaces subscriptions, agents will make more buying decisions than people, and products need machine-readable prices.

View example

Both carry commercial disclosures behaviorally as well: one closes with "Hire viktor.com for your team. $100 in credits included, no card." He frames the hire-vs-build decision — "proprietary logic, build it. team-wide reliability, hire it."

Original disclosure

Sponsored post tagged "#sponsored"; the partnerships post tagged "Paid Partnership".

View example

A domain returns the site's CMS, analytics, payment tools and ad pixels; a technology name returns the websites using it — useful, he notes, for targeting sites that run a given tool.

• Each call costs about 5 cents, with a free call showing tech categories.
• He says the partnership was signed by looot so users don't open accounts themselves.

View example

A name and company return a verified work email with job title and work history; other calls search people or companies by title, seniority, industry or headcount, and enrich a company from its website. Each call costs about 4 cents with no charge on no-match, and adding a mobile number takes a lookup to about 39 cents.

View example

No commercial terms are given; the post presents it as a product integration rather than a sponsorship.

View example

In the same post he says he gives demos of "our semi-autonomous departments and companies," and elsewhere writes "I build gtm x ai systems for a living." A separate post about which GTM motions pay off cites "our GTM OS data" alongside a survey of 100+ GTM experts, tying the product to his own research.

View example

He is commenting on the company and market, not promoting or investing in it; no commercial relationship is stated. Claude tools also appear in his posts as the agents most of his looot customers call from.

View example

Gannon's post about hiring Viktor to check inbound leads, update the CRM and give time back is the same buy-an-agent-for-a-defined-task pattern Arnoux describes in his revenue operations and partner outreach case studies.

Gannon's posts are shorter first-person founder anecdotes, including a brand-relationship story about a personalized Circle jersey. Arnoux builds larger frameworks, channel economics and infrastructure arguments, and often uses diagrams and launch metrics.

Read original · Original 2

Moura's 7-stage AI ladder and workflow-versus-agent post ask the same kind of diagnostic question Arnoux uses, such as whether a system is handed steps or a goal.

Moura teaches an adoption ladder and simple conceptual distinctions in plain list posts. Arnoux focuses more on enterprise operating models, agent data pricing, GTM channel selection and infrastructure, with more detailed diagrams and commercial examples.

Read original · Original 2

Kirk's hidden-CRM-audience story and runaway-workflow story match Arnoux's interest in reliability, buying decisions and system guardrails.

Kirk writes short problem-and-lesson posts about intent data, CRM hygiene and automation trust. Arnoux covers a broader terrain of AI operating models, agents, provider economics and compute, often with formal frameworks.

Read original · Original 2

The reviewed Viktor post shows Gannon checking inbound leads across email and booking forms, adding missing leads to the CRM, building a monthly lead chart, and reporting about two hours a day back since July. That is a concrete, personal account of a ready-made AI employee doing one job, which suits a sponsor selling a lightweight agent that a founder or small agency can adopt quickly.

The tradeoff is decision depth.

• Gannon's reviewed material gives one operator's time saving and a simple lead-handling workflow, but it does not give the build-versus-buy test, the overnight pipeline review, the forecast-gap example, or the compliance boundary in Arnoux's posts.
• A sponsor that needs RevOps leaders to weigh architecture, governance and buy-in will get a stronger case from Arnoux.
• A sponsor that needs a relatable founder saying the tool paid for itself in hours will get a stronger case from Gannon.

Examples: Source 1 · Source 2

The reviewed posts map seven stages of AI use at work and draw a clear line between a workflow that follows written steps and an agent that receives a goal.

• Moura uses an accounts payable example to make the distinction concrete.
• That structure gives a sponsor a self-assessment hook — which stage is your team on? — and a vocabulary for agent adoption that a general business audience can follow without a product tour.

The tradeoff is commercial and operational specificity.

• The reviewed Moura posts are educational and framework-led; they do not carry a sponsorship or paid-partnership label in the supplied material, and they do not show a GTM pipeline, forecast review, vendor pricing or a compliance approval boundary.
• A sponsor using Moura would need to supply the product link, campaign mechanics and deployment proof.
• Arnoux already demonstrates those sponsored mechanics and supplies a RevOps workflow story, so Arnoux is stronger when the campaign must show a buyer deciding on a specific agent purchase.

Examples: Source 1 · Source 2

The reviewed posts show a client about to spend roughly $30,000 on intent data before finding 50,000 people already in the CRM who had never been followed up with, and a client with more than 100 workflows where one sent an old test email to 600 people.

• Both examples give a marketing-ops audience a reason to clean up and trust the existing system.
• A sponsor selling data hygiene, workflow governance, or CRM audit services fits that task more closely than an agent-platform message.

The tradeoff is that Kirk's reviewed material does not cover build-versus-buy agent decisions, agent architecture, or a documented sponsored format.

• For a sponsor selling an AI agent or RevOps platform, Kirk would supply the diagnosis — stale contacts, unexplained workflows, eroding trust — but not the agent selection argument.
• Arnoux supplies the agent decision frame and a deployment example.
• Kirk is stronger when the audience's first problem is the data and automation already in place; Arnoux is stronger when the audience is choosing an agent to buy or build.

Examples: Source 1 · Source 2