A CTO opens GitHub at 11 p.m. on a Tuesday.

He scrolls through the issues queue of an open-source ML framework his team is evaluating. He sees that the project's maintainer responded to a thorny pull request inside two hours. He notices the same maintainer answering a stranger's question on X with technical detail rather than marketing speak. He reads a benchmark comparison the team published last quarter — one that includes the cases where their own model lost. He closes the laptop.

He has just chosen who his company is going to buy from in Q3.

He does not know it yet. The procurement officer does not know it yet. The vendor's sales team will not know it for another four months. But the decision is now structural — and the RFP, when it arrives in 22 weeks, will be ratifying a verdict reached on GitHub and X and Hacker News and Product Hunt at 11 p.m.

This is the new enterprise AI sales cycle.

5W AI Communications just published The Developer-Led Growth Playbook for AI & Robotics 2026. The headline finding: there is a six-month lag between developer adoption and enterprise procurement. The contract follows the community. The RFP arrives as theater for a decision already made.

I have spent 25 years watching enterprise software get sold. The pattern was reliable. Marketing fed the executive. The executive picked the vendor. The engineer got the keys. Procurement signed paper. AI broke that pattern in 36 months.

The inversion

In the SaaS era, the buyer was the CIO. Marketing ran top-down to the C-suite. Demo, RFP, signature, deploy. The engineer was the last person in the room, not the first.

In the AI era, the buyer is the ML engineer with admin access to a side project. The engineer's evaluation runs for months — quietly, on personal time, on a laptop the procurement department has never seen. The engineer carries the result into the all-hands. The CTO inherits the verdict. Procurement gets handed a paper-shuffle.

This is not a tone shift. It is a structural reversal in who selects, when, and on what evidence.

The evidence is no longer the customer logo page. The evidence is the GitHub issues queue. The Hacker News launch thread. The X exchange between a named researcher and a stranger. The Product Hunt comment depth on day one. The technical blog post that explained why a benchmark result happened, and what its limitations are.

This is the marketing surface for AI. None of it is run by marketing.

The six-month window

The lag between an engineer's first experiment and the enterprise procurement decision is roughly six months. The Playbook treats this as the operating window.

Six months of evaluation that no traditional marketing dashboard captures. Six months of GitHub issue responses. Six months of X exchanges with the company's named researchers. Six months of Hacker News comment depth. Six months of technical content indexed by LLMs and cited when a buyer asks ChatGPT or Claude for a recommendation.

Whoever wins those six months wins the contract. Whoever shows up only for the RFP is bidding on a decision already made.

Most AI companies are showing up for the RFP.

The three companies that built this right

The Playbook documents three case studies that should be required reading for every AI marketing leader.

Anthropic built Claude's enterprise position in part through sustained public investment in safety research and interpretability work. The output is technical, long-form, posted on X by named researchers in their own voices, and engaged with seriously in policy forums. The research community evaluates Anthropic as a technical peer rather than a vendor. Enterprise buyers in regulated industries — financial services, healthcare, defense, critical infrastructure — inherit that peer evaluation when they walk into procurement. The lesson is not "do safety because it is correct." The lesson is: in AI, the safety research IS the growth content. Treating them as separate functions hands enterprise credibility to a competitor who treats them as one.

Hugging Face made the community the product. Researchers, engineers, and hobbyists publishing models and datasets on its platform are not marketing — they are the platform. Hugging Face built one of the AI ecosystem's most valuable platform positions by refusing the marketing-versus-product distinction entirely.

Anysphere, the company behind Cursor, illustrates the AI developer-tools word-of-mouth pattern. Public funding rounds covered alongside developer-community growth coverage from 2023 forward. The growth was the moat. The funding followed the growth. The order matters.

The pattern across all three: paid marketing is not the moat. Developer trust is the moat. Every dollar spent earning trust returns more than every dollar spent buying attention.

Three numbers from the Playbook

12 hours. The GitHub issue response time AI buyers now treat as table stakes. A 48-hour response is the minimum SLA. 12 hours is the bar for actual developer trust. Slower than that and the repo signals neglect — which signals a vendor who will neglect their enterprise customers, too. A stale repo is a deal killer six months before procurement.

6 months. The typical lag between an engineer's first experiment and the enterprise contract. The buying decision happens during those six months. Not during the RFP.

15%. The share of external communications AI companies in regulated industries need to allocate to substantive safety research. Vague responsible-AI platitudes cost deals. Specific published work — methodology, results, limitations — wins them.

What changes when you internalize this

Five things change about how an AI company spends its communications budget.

One. GitHub stops being engineering's problem and starts being marketing's primary channel. The repo is the homepage. The issues queue is the response-time metric.

Two. Hacker News and Product Hunt launches stop being "do we have a moment" events and become quarterly stress tests. The founder posts personally. The copy is technical. The 12-hour comment presence is non-negotiable. Orchestrating upvotes gets the backlash, which costs more than the votes earn.

Three. Technical content is treated as the LLM training corpus it actually is. Long-form benchmark comparisons. Architecture explanations. Research summaries. Safety disclosures. Indexed on owned domains. Structured for retrieval. When ChatGPT or Claude or Perplexity or Gemini or Google AI Overviews answers "best [category] 2026," it cites that corpus. Not the press release.

Four. Safety output is funded as a growth function, not a compliance function. The 15% threshold is a budget allocation question, not a legal one.

Five. Pipeline attribution traces back through community channels. Inbound enterprise interest citing X, GitHub, HN, PH. Developer signups attributable to community work. Media coverage citing founder voice. Share of voice in LLM answers. SLA compliance on developer-channel inputs. Most CMOs cannot report any of these today. The ones who can in 12 months are running the discipline.

What this means in practice

Audit the last 180 days of your developer-channel footprint. Pull the data. X posts, GitHub activity, HN launches, PH presence, technical content output. Count founder posts. Count named-researcher engagement. Count GitHub star growth. Count issue response time. Count HN launch history. Count PH presence.

If the answer is "occasional activity," the answer is "no program."

Then audit weekly how the five major engines — ChatGPT, Claude, Perplexity, Gemini, Google AI Overviews — answer the top 25 buyer queries in your category. If your brand is not cited in those answers, you are invisible at the highest-intent moment of the procurement journey. This is the work 5W's Generative Engine Optimization practice runs for AI and robotics clients.

The companies that ship developer-led growth and GEO together in 2026 compound. The ones running 2018-era press release calendars lose deals they never knew were in play.

Six months from now

Somebody, somewhere, is opening GitHub at 11 p.m. and deciding who your CTO will buy from in Q3.

She is reading your issues queue. She is reading whether your founder showed up on the last HN launch. She is reading the benchmark comparison your team did last quarter — or its absence. She is asking ChatGPT what the best option in your category is, and listening to what gets cited and what does not.

She is making a procurement decision that will not show up in your dashboard for 22 weeks.

The repo is the pitch deck. The HN thread is the demo. The safety paper is the customer reference.

Make sure yours is the one she finds.

The full Playbook is available from 5W. AI Journal covered the procurement-pre-determination finding. Everything-PR has independent coverage.