Picture the enterprise software sale of the near future: no human ever reads your developer docs. Instead, an AI agent scrolls through your API reference, evaluates code samples against live endpoints, and decides—autonomously—whether to integrate your product. The sale closes without a demo call.
That vision, still speculative for most companies, is the wager behind Manicule, a San Francisco startup trying to invent what it calls "AgentRel"—developer relations rebuilt for software that pitches itself to machines, not people.
Fresh from Y Combinator's Spring 2026 cohort, the five-person team has clawed its way to $58,000 in monthly recurring revenue, claiming 100% month-over-month growth across recent months. Clients include a clutch of AI-native startups: Supermemory, Greptile, Reducto, Rootly. These aren't household names yet, but they share a common problem. If autonomous agents really do start choosing which APIs to call, traditional documentation—written for tired developers skimming Stack Overflow at midnight—won't cut it.
"We write for agents and humans," cofounder Naman Bansal said during a late-May podcast, describing what he frames as more than a documentation refresh. It's closer to a studio model: AI-driven audits of existing docs, paired with human writers who understand information architecture. The humans still lead. The agents verify.
Manicule runs its own GitHub app that crawls client repositories hunting for gaps—outdated code snippets, missing endpoints, vague capability descriptions. Then comes the verification layer: agents test every code sample against actual APIs, flagging what breaks. Human editors shape the structure and tone. The company says it ships revamped docs in around 15 days, faster than most engineering teams can schedule the internal meeting to discuss scheduling a docs refresh.
Positioning Between Platforms and Promise
The pitch lands Manicule in contested territory. Documentation platforms like Mintlify let companies host AI-optimized content, but you still need a team to write it. Answer engine optimization tools measure how visible you are to AI systems scraping the web; they don't produce the content itself. Manicule wants to own the whole pipeline, a service play rather than self-serve software.
According to the company's Y Combinator launch materials, clients have seen support ticket volume drop roughly 37% after implementation. Supermemory, one early customer, reportedly logged a 37% bump in answer success rate and closed "multiple enterprise deals directly from docs." Those numbers come from Manicule's own reporting—no independent case studies have surfaced yet, and client endorsements remain confined to company-controlled channels.
Still, there's a logic to the model, particularly for startups without dedicated DevRel headcount. If you're a 12-person team building AI infrastructure, hiring someone to rewrite docs for agent consumption isn't just expensive—it's a weird skill set to recruit for. Manicule sells itself as the outsourced answer.
The team is hiring aggressively, or at least ambitiously. A recent job listing for a "Content Engineer" asks for fluency in generative engine optimization and the ability to "build your own agents for fun." The posting also promises to cut blog turnaround from the current 3.5 days to one day, an aggressive timetable that either signals confidence or a misunderstanding of how bottlenecks actually work in content production.
Naming Rights to a Category That May Not Exist Yet
By branding its approach "AgentRel," Manicule is doing something shrewd and slightly desperate: trying to name a market before anyone agrees the market exists.
The term "Agent Relations" has bobbed up in DevRel circles and enterprise governance discussions over recent months, but there's no formal category yet, no Gartner quadrant. It's a land grab that echoes earlier B2B plays—"customer success" didn't exist as a job title until SaaS companies willed it into being, then fought to own the definition. Same with "growth marketing." Name the function, build the budget justification, hire the team.
Whether AgentRel becomes real depends on a bigger question: how fast do AI agents actually start making purchasing decisions? There are signals. New Relic launched an agentic platform in February. Kore.ai rolled out its Artemis agents in May. Governance vendors like Entro and Noma introduced agent access controls in March and June, respectively. In March, a startup called AgentMail announced a $6 million raise to build email infrastructure specifically for bots talking to bots.
The scaffolding is going up. Manicule wants to be the layer that gets agents to discover your product in the first place, the top-of-funnel play in a world where funnels run on code, not persuasion.
What Remains Murky
Pricing remains undisclosed. Given the studio model and the named clients—mostly VC-backed AI startups—it's reasonable to guess deals run into five figures annually, maybe low six if the scope expands. CB Insights data from late May indicates a $500,000 convertible note, though Manicule hasn't formally announced or confirmed funding. (Convertible notes are the duct tape of early-stage finance; they delay the valuation argument until a proper round materializes.)
Customer validation is thin. The clients Manicule lists are real companies, many from the YC network, but none have gone on record with independent endorsements. For a service promising to unlock enterprise deal flow through better documentation, third-party case studies would matter—especially as Manicule tries to sell beyond the YC alumni club.
And then there's the central uncertainty, the one Bansal and his team can't control: timing. If autonomous agents begin driving vendor selection quickly—say, within the next 18 months—then early movers in agent-optimized content will enjoy a structural advantage, the same way early SEO-focused companies dominated search a decade ago. If adoption drags, traditional documentation playbooks may simply evolve incrementally. Add an llms.txt file here, tweak some metadata there. Good enough.
Manicule is betting the shift won't be incremental. That the best way to reach an AI agent isn't through crawling protocols or schema markup, but by fundamentally remaking how developer content gets created in the first place—built for machine readers from the start, with humans in the loop to keep it from reading like a spec sheet.
Perhaps the most revealing thing about Manicule's approach is what it assumes about the future: not just that agents will make decisions, but that those decisions will hinge on documentation quality. As if, in a world of autonomous software, the pitch still matters. Only now, you're pitching to something that doesn't get bored, doesn't skim, and never closes the tab to check Slack.
It just reads. And decides.
