Lapo Chirici

B2B marketing meets AI — for real, not for slides.
CEO @Krein_it · AI Marketing Alliance · Author @SpringerNature · Building Krein Tools

🔗

[Lapo Chirici @Lapo_]

https://t.co/oPuJhJaPwz

[rvaniaaa @rvaniaaaa]

This is f*cking gold, the CEO of Obsidian put his own second brain on GitHub for free and people still build theirs from zero

7 free repos from the Obsidian team:

-> kepano obsidian (by kepano)
https://t.co/lT5QtPucOn
the vault template of the Obsidian CEO himself, copy it and skip the blank page

-> jsoncanvas (by obsidianmd)
https://t.co/iNkNYs8Y7k
the open canvas format, your maps stay plain files that are yours forever

-> obsidian releases (by obsidianmd)
https://t.co/AMjE6dW82S
every community plugin and theme in one list

-> obsidian sample plugin (by obsidianmd)
https://t.co/Ytp6e1mcDo
the official template, build the plugin you are missing

-> 40 questions (by kepano)
https://t.co/3BJ35OEFsA
40 questions that turn your notes into a yearly review

-> obsidian minimal (by kepano)
https://t.co/idJC0tBWMJ
a clean theme so your vault stops looking like a mess

-> flexoki (by kepano)
https://t.co/FswpE9XrZ4
the color scheme he made for writing and code

your second brain should never start from a blank page

copy his vault and make it yours

[Ruben Hassid @rubenhassid]

https://t.co/ZOEVRy8Sok

[Sprytix @Sprytixl]

ANTHROPIC LEAKED A FILE WHERE AN AGENT WITH NO CODE BUILDS A SYSTEM WORTH $100,000 IN 60 MINUTES

the agent costs $0, builds the whole thing locally with no code at all, and month one is already $70K+

services → form → checkout → mailer → cash

it writes nothing - it opens the 6 services you already pay for and configures them itself

at minute sixty there is a live page, a form writing to a sheet, a link that takes money and a sequence that sends itself

no repo, no tests, no diff and no rollback - and that is the whole speed and the whole risk

$100,000 goes in and $96,883 lands in the bank: the checkout takes $2,996, the six tools $120, the agent nothing

the checkout eats thousands of times more than everything the agent did in that hour

the agent closes three quarters of the circle by itself - it builds, it publishes, it waits

you close the fourth quarter, and that is 334 people at $299 the hour never brings

a coded build leaves you something you can read a year later, this one leaves you six accounts

save this and paste it into Claude Code - let it configure these 6 services for you ↓

[Sprytix @Sprytixl]

ANTHROPIC + GROK LEAKED A FILE WORTH OVER $4M THAT BUILDS A SOLO AI BUSINESS DOING $100K A MONTH IN 60 MINUTES

the hour costs $2.20, and day one can close $800

root → reality → surface → hunt → clock → back to root

5 blocks of 12 minutes, and two of them eat 68% of the bill because only they touch the live feed

Grok 4.6 hands over not a feeling that there is demand, but 4 named fields: who is already doing it, what they offer, what people complain about and where the hole is

the fourth field decides what every block after it builds

Opus 5 gets called exactly 1 time, and every later block quotes it

3 Sonnet 5 run in parallel and produce 4 files in 12 minutes, including the payment link most launches are still missing on day 30

400 signals → 12 letters → 3 replies → 1 payment, and this is the only block that produces a conversation

the hour buys you a page, a link, a queue and a signal list - it does not buy you buyers

save this and paste it into Grok - let it build your ROOT.md ↓

[Sprytix @Sprytixl]

ANTHROPIC LEAKED 10 TOOLS THAT BUILD A PRODUCT IN 24 HOURS - AND ALL OF THEM ARE FREE

everyone clones the 10 repos, almost nobody stacks them in the right order - the order is the product.

core → scaffold → work → gates → autonomy

the first layer isn't writing code - it's assembling a team: the core in the terminal, procedures as files, plugins on top.

you clone 3 repos and you've hired nobody, yet 6 roles already exist.

the scaffold is the biggest saving on the whole path - you don't design twice what's already solved in the templates.

the patterns sitting in the cookbooks are the exact ones you'd rediscover on your own in week 3.

then the work runs in a room you were never in - an agent where thinking is needed, a direct call where it isn't.

gates replace review, they don't supplement it - the agent sits inside your pull request and rejects before you open it.

a gate that never rejects anything is a status badge, not a gate.

only the last of the 5 layers is autonomy - scan, research, patch, and one report waiting for you while you pressed nothing.

turn autonomy on before the gates and you've just automated your own mistakes.

3 inversions in the order and the same 10 repos produce zero product.

save this and paste it into Claude Code - the repos are the public part, the sequence isn't ↓

[Lapo Chirici @Lapo_]

@jack_9947 AGENTS

[Jack J. @jack_9947]

All 50+ B2B SaaS Agents (Free for First 300 People)

CLAUDE AGENT DIRECTORY FOR B2B SAAS - FREE

1. Sales, Prospecting, and CRM (14 agents)
2. Outbound and Cold Email (8 agents)
3. Marketing and Growth (10 agents)
4. Product and SaaS (8 agents)
5. CRM, RevOps, and Finance (6 agents)
6. Customer Success and Retention (5 agents)
7. Finance, Metrics, and Strategy (4 agents)
8. Multi-Agent Frameworks and Orchestration (5 agents)

(72 Hours only)

Like + RT + comment 'AGENTS'

Must Follow me so I can DM you.

[Jack J. @jack_9947]

All 17 Outreach Skills (Free for First 300 People)

B2B OUTREACH PLUGIN - FREE

1. Onboarding Skills, triggered via /nl-outreach:onboard (8 skills)
2. Copywriting Skills: cold email, LinkedIn DM, cold call
3. Analysis, Pipeline, and Automation Skills
4. The Persistent Brief, updated once, read by every skill after

(72 Hours only)

Like + RT + comment 'BRIEF'

Must Follow me so I can DM you.

[Lapo Chirici @Lapo_]

https://t.co/V8Rr6nuGPG

[Lapo Chirici @Lapo_]

Forrester: B2B agency spend on digital dropped 51% → 31%, content 41% → 26%, year over year.

Everyone's calling it budget cuts.
The real shift: AI changed the make-vs-buy threshold.

[Lapo Chirici @Lapo_]

For B2B operators: the question isn't "do we still need agencies?"

It's "which parts of the funnel have an AI tax on coordination, and which still compound through human asymmetry?"

Automate the layer. Invest in the edge.

[Lapo Chirici @Lapo_]

https://t.co/C2ts1wBR7Z

[Lapo Chirici @Lapo_]

Demand gen used to run on signals you could instrument: form fills, ad clicks, webinar registrations. That machinery is breaking because buyers now conduct discovery inside LLM interfaces that produce no MQL-shaped exhaust. The question isn't whether to optimize for AI search — Google shipped AI Overview enhancements in May 2024, OpenAI launched ChatGPT search in October, and Microsoft embedded Copilot across Bing. Multiple frontier vendors converged on the same primitive within eight months. The question is what signal architecture replaces the one you built.

G2's playbook names three layers most demand gen teams are conflating. Layer one: corpus legibility — can an LLM surface your brand when it's never indexed third-party validation, case studies, or comparison data? Layer two: citation gravity — do the sources LLMs retrieve (reviews, community posts, technical docs) actually name your category and your product in the same breath? Layer three: agent-ready infrastructure — can a buyer's AI agent complete discovery, vendor comparison, and intent capture with zero form friction?

For B2B demand gen leaders managing 2026 planning cycles, this isn't an SEO refresh. It's a re-instrumentation problem. The old funnel assumed you controlled the first touch and could track every subsequent signal. The new surface assumes the LLM mediates discovery, your brand shows up only if structured data and third-party proof already exist in its retrieval set, and intent signals arrive late or not at all. If your demand gen stack still measures top-of-funnel by gated content downloads and bot-filled webinars, you're optimizing the wrong gate.

Vendors win not by gaming prompts but by ensuring that when an LLM constructs a shortlist, verified proof already exists in the corpus it searches. That's why review density, community conversation volume, and machine-readable case studies are now demand gen infrastructure, not marketing collateral.

The frontier isn't visibility. It's retrieval-worthiness.

[Lapo Chirici @Lapo_]

https://t.co/4O0KAufAcP

[Lapo Chirici @Lapo_]

Can an AI agent read your product page? Not "can it rank it"—can it extract the spec table, parse the pricing tiers, cite your company in an answer? If your site relies on client-side rendering, the answer is no. And the gap is already rewriting traffic composition.

One Contentsquare client—a financial services brand—lost 40% of organic search traffic to a single page. Conversion on that page increased. AI-referred traffic grew 632% across Contentsquare's client base in ten months. The visitors arriving from ChatGPT Search, Claude, and Perplexity are more informed, more intentional, and fewer in number. But most brand sites can't see them coming, because the bots see nothing.

Here's the structural problem: most AI crawlers fetch raw HTML only. No JavaScript execution. No headless browser. Client-side rendering sends them an empty shell. Google's crawler is the exception—it runs a headless Chrome service—so a site can rank in Search while being invisible to every AI answer engine. OpenAI, Anthropic, and Perplexity all moved on the same primitive this quarter: raw HTML extraction, bypassing render layers entirely.

The fix is server-side rendering. Every product page, comparison table, FAQ section, pricing grid must be assembled server-side and shipped complete. That's table stakes. The second layer is dual-mode media: transcripts for videos, alt text for visuals, structured metadata for interactive modules. AI agents don't watch a demo; they read its transcript. If the transcript doesn't exist or doesn't describe what the video shows, the content is invisible.

B2B brands are accidentally ahead. Spec sheets, comparison tables, FAQ sections—content built for procurement committees—parses cleanly for AI agents. Software buyers are increasingly beginning their research in AI-powered tools, and B2B content structures align naturally with how these systems extract and process information. B2C brands relying on rich visuals and minimal text are structurally behind.

For CMOs and RevOps leaders evaluating 2026 web infrastructure: the question isn't whether your site ranks. It's whether an AI agent can extract, cite, and route buyers from it. Rendering is now a go-to-market decision, not a dev-ops footnote.

Different surfaces, different mechanics, different optimization.

[Lapo Chirici @Lapo_]

Anthropic, OpenAI, Microsoft, and Google have all shipped the same primitive across the last eight months: MCP, GPT integrations with governance layers, Copilot controls, Workspace automation with data boundaries. Four labs converging on centralized governance with decentralized execution. The pattern most B2B operators are missing isn't the protocol race. It's the shift in where GTM tooling actually lives.

Clay's MCP for reps lands inside Claude, ChatGPT, Microsoft 365 Copilot, Glean — wherever the rep already works — while Ops sets workflow permissions, caps credits, and controls which records write back to the CRM. The job-to-be-done isn't giving reps another dashboard. It's letting them prospect in the surface they've already adopted, with the data layer Ops still owns. The governance stays infrastructure-grade. The interface stays rep-native.

For B2B SaaS vendors still building dashboards as destinations, this is a category rewrite signal. If Ops can govern Clay workflows inside Claude the same way they govern Salesforce permissions, the differentiation isn't your UI anymore. It's whether your data layer speaks MCP, whether your governance model can follow the rep into ChatGPT, and whether you understand that "centralized control, decentralized surface" is now table stakes — not a roadmap item.

Spider's free local crawler makes the shift concrete: discovery happens in a tool the rep controls, enrichment costs hit only the records that passed triage, and the CRM sees clean net-new writes instead of duplicate slop. Same job as the $200/seat signal platform, different boundary.

The vendors that survive this aren't the ones with the best dashboard. They're the ones whose data moves.

[Lapo Chirici @Lapo_]

Can an AI copilot pay for itself if the rep still waits three weeks for the buyer's legal team? Most GTM leaders can't answer that with data.

Sales AI spend in 2026 has become a line item without a P&L line of sight. Millions flow into conversation intelligence, coaching agents, CRM automation, pipeline generation. Yet when boards ask "Is this working?" most teams point at adoption dashboards, not revenue impact.

The metric that matters is Productivity per Rep (PPR): new revenue generated per AE over a defined period. Not seats activated. Not prompts run. Revenue per head.

PPR breaks into four variables: number of opportunities × close rate × ACV ÷ sales cycle length. AI should move at least one of those levers without degrading the others. Here's where the attribution breaks down.

ACV is the wrong lever. Pricing reflects positioning and product differentiation, not GTM tooling. If AI reduces CAC, pass the savings to buyers—don't inflate prices.

Close rate and cycle length are more promising but buyer-dependent. A rep can execute flawless discovery, generate a perfect follow-up email, customize a demo in five seconds, and still wait two quarters for procurement. The bottleneck isn't your motion—it's their committee.

The highest-leverage variable is opportunity volume. AI can compress prospecting, qualify faster, surface intent signals, automate outbound sequences, and route inbound at machine speed. This lever is seller-controlled, measurable, and compounds.

But volume without quality is a tax on CS and a retention time bomb. The quality gate is Leading Indicator of Retention (LIR), tracked by cohort and by AE. Speed gains mean nothing if discovery quality degrades, win rates drop, or retention suffers months later.

That's the test most teams aren't running. PPR without LIR measures throughput, not value. For B2B operators evaluating 2026 AI investments, the framework is simple: instrument PPR, guard it with LIR, and isolate which RVF variable your tooling actually moves. If you can't measure the delta, you're not buying infrastructure—you're renting hope.

Productivity without retention is just expensive churn.

[Lapo Chirici @Lapo_]

https://t.co/PyhkyOXmfx

[Lapo Chirici @Lapo_]

Can an AI agent extract your pricing without defaulting to Reddit or G2? Most B2B vendors can't answer yes.

Siteline ran Claude agents across 100 B2B products with three buyer tasks: pricing, integrations, security. The pattern was clear: pricing consistently underperformed. When agents couldn't parse or cite a vendor's own page, they pivoted to third-party sources—even when a numeric price was published.

The failure isn't access blocking (7% of runs). It's machine-readability: pricing buried in JavaScript calculators, toggle menus, PDFs, screenshot images, or ambiguous tables. Humans see the page fine. Agents can't extract, can't cite, can't trust the structure.

OpenAI, Anthropic, and Microsoft all shipped agent frameworks with enhanced web navigation and structured fact-extraction this quarter. Google introduced agentic tasks in Search. Salesforce reported 20% of sales now sourced from agents. The primitive converging across labs is the same: agents that fetch, parse, and cite—not browse.

For CMOs and demand-gen operators, this is a compression cycle. If your pricing page fails the agent parse test, your prospect's buying agent cites a competitor or a review site instead. You're not losing traffic; you're losing attribution and control at the moment of highest intent.

The fix isn't SEO. It's structured discoverability: semantic HTML, clean tables, https://t.co/bVCKBV71fk markup, and machine-readable pricing logic. The agents your prospects deploy in 2026 won't ask for a demo if they can't parse your tier grid.

Agents don't read showrooms. They scan barcodes.