This is the first month of tracking Romain's feed, so there is no prior baseline to compare against. Read on its own terms it was a confident start. Nine posts pulled 12,258 impressions and grew the following by 281 to cross 11,000, with clean, steady acceleration into the final week around Glean:GO. The feed is tightly on-message: nearly every post argues one coherent thesis, that the integration and tools layer, not the model, is where enterprise AI actually succeeds or fails (MCP maturity, token spend, shadow AI, ROI measurement). Reach was led by two very different posts: the partner-ecosystem update welcoming TYME and Enboarder (2,634, the month's widest reach) and the "MCP is like a toy today" hot take (1,934). The Glean:GO keynote recap earned the strongest engagement rate of the month at 5.5%. The one clear gap is saves: just 6 across the month, and only 3 posts earned any at all. The frameworks, which are the format a technical audience normally bookmarks, reached the lowest of any pattern and produced only 2 saves between them. The thesis and the voice are landing. The opportunity for September is to package the deep-technical posts so buyers keep them, and to convert high engagement into durable intent.
+281 net followers, from 10,785 to 11,066, the month the account crossed 11,000. Growth was consistent all month with a visible step up in the final week (Aug 27 to 31) around the Glean:GO keynote and its recap.
A healthy spread. The partner update led at 2,634, four posts cleared 1,400, and the tail sat in the 750 to 950 range. No single post ran away with the month, and no post flopped.
Each list ranks by a different signal: impressions for reach, saves for buyer intent, and engagement rate for how hard a post worked relative to its reach. The notable split this month: the two widest-reach posts (partner update, MCP hot take) were not the two hardest-working, and the highest engagement rate belonged to the Glean:GO recap, which reached solidly but not widest.
Who got seen the most.
Saves are the strongest B2B intent signal. This month they were thin: only 3 posts earned any, each with 2, and the framework posts under-converted despite being the natural bookmark format.
How hard each post worked for the reach it got. Reactions, comments, shares, saves, and sends over impressions.
Image-led: four single-image posts, two multi-image, two text, and one article (the partner update). No video shipped this month.
The single article (partner update, 2,634) tops the chart, but that is one post, so treat it as directional. Image, multi-image, and text all landed close together between 1,000 and 1,300. Format is not the driver this month; topic and hook are.
We read each of the 9 posts and classified them by the job the post was doing, adapted to StackOne's audience of technical buyers, platform teams, and IT and security leaders evaluating an enterprise integration and tool-governance layer for AI agents. Five patterns appeared. The month leaned on hot takes and educational thought leadership: five of nine posts argued a point of view or taught a concept.
The single product update led on average reach, followed by hot takes and the event posts. Frameworks and the lone founder-arc post trailed, both landing under 810 average impressions.
Engagements (likes, comments, shares, saves, sends). Hot takes drove the most total interaction at 147, with the two event posts close behind at 133, carried by the Glean:GO recap.
Five content patterns across 9 published posts. Product update and founder arc had only one post each, so treat their averages as directional.
| Pattern | Posts | Total imp. | Avg imp./post | Total eng. | Saves | Top performer |
|---|---|---|---|---|---|---|
| Hot take / POV | 3 | 4,592 | 1,531 | 147 | 2 | MCP is like a toy today (1.9K imp) |
| Event promo / recap | 2 | 2,673 | 1,337 | 133 | 0 | Glean:GO keynote wrap (1.8K imp) |
| Framework | 2 | 1,600 | 800 | 65 | 2 | MCP Evolution (832 imp) |
| Product / company update | 1 | 2,634 | 2,634 | 63 | 2 | Welcome TYME & Enboarder (2.6K imp) |
| Founder arc / journey | 1 | 759 | 759 | 31 | 0 | Romain's usage logs (759 imp) |
The three hot takes averaged 1,531 impressions, the highest of any repeatable pattern, and drove 147 of the month's 439 engagements (33%). The strongest, "MCP is like a toy today" (1.9K, 2 saves), worked because it took a contrarian position and then backed it with named, concrete failures: QuickBooks MCP that could not handle dual entities, Jamf's official MCP that dropped required fields. That is the formula to keep mining: a sharp claim about the integration layer, proven with specific vendors and real enterprise failure modes rather than abstractions. It is combative in the right way and it always routes back to why StackOne exists.
Frameworks are the format a technical audience bookmarks, yet the two this month ("Token spend is a tools-layer problem" and "MCP Evolution") averaged just 800 impressions, the lowest of any repeatable pattern, and earned only 2 saves between them. The content is genuinely save-worthy (a 3-mechanism token-efficiency model, three operational MCP pains) but it was packaged as dense prose without a clean, numbered, steal-this takeaway up top, and both shipped early in the month before the follower base had grown. The single highest-leverage change for September is to keep the framework substance but package it for the bookmark: numbered, scannable, with the payoff promised in the first two lines.
The Glean:GO keynote wrap earned the month's best engagement rate (5.5%) on strong reach (1.8K), well ahead of the earlier "where we'll be" logistics tease (912). The recap worked because it carried a thesis, the read-vs-write gap for AI agents, not just a thank-you. The lesson: event posts earn their slot when they argue something. Favor recaps that make a point over pure teases, and tie every event moment back to the core integration argument.
The first sentence decides reach on LinkedIn. Romain's hooks are already strong, and the pattern is consistent: the posts that led on reach and engagement stacked the most of these five ingredients. We analyzed the opening lines of the month's highest-reach and highest-resonance posts.
"only reached about 15% of their potential capacity" (ROI post), "deploy AI to 1000s of employees and then panic at the bill" (token spend), "1,061 hours of human-equivalent work over 90 days" (usage logs). Concrete numbers in the opening set a stake and signal substance. The posts that opened on a hard figure consistently earned the strongest engagement rates.
QuickBooks and Jamf as concrete MCP failures, TYME, Enboarder, Glean, and Workday in the partner update, Claude in the shadow-AI and usage-log posts. Real names carry proof and pull in adjacent networks. The two widest-reach posts of the month (partner update, MCP hot take) both leaned hard on named entities.
"We are currently in the middle of an MCP bubble", "you can't block your way out of shadow AI. The more you restrict, the more of it you create", "MCP connectors are nowhere near ready out of the box". Each hook sets up a contradiction the reader wants resolved. This is Romain's strongest and most native move, and it drove the hot-take pattern to the month's best average reach.
"Here's why neither one solves the problem", "Here's what he found", "Here's why". The posts that promised a concrete payoff in the opening pulled readers into the body. This was applied unevenly: the strongest teaching posts made the promise, but none framed the payoff as a numbered, bookmarkable list, which is the direct tell for why saves stayed thin.
"I talk to teams every single day who hit this exact wall", "I chatted with an enterprise CFO the other day", "I've spent months watching teams deploy AI". Romain writes from constant, direct contact with buyers, and that vantage is what makes the takes credible rather than generic. Every top post this month was anchored in a specific, first-person conversation or observation.
Quality was high across the board and there were no misfires, so the opportunities are about reach and conversion rather than failures. Two posts sat lowest on reach, and both point at the same fix: package deep-technical substance so it travels and gets kept.
This was a genuinely interesting founder experiment (asking Claude to quantify ROI from StackOne's own usage logs) and it earned a healthy 4.1% engagement rate, so the people who saw it responded. Reach was the problem, lowest of the month at 759. Two likely factors. First, it shipped Aug 7, early in the month before the follower base had grown, so it started from a smaller base. Second, the copy repeatedly hedged its own authority ("I one-shotted this", "unproven", "it'll need a bit more work"), which softens the signal and gives the reader a stat dump rather than one clear, quotable conclusion to react to or keep. The underlying insight ("token usage measures activity, tool calls measure work") is excellent and deserved to be the hook.
This is the best framework of the month (a clean 3-mechanism model: strict payload filtering, progressive tool disclosure, architecture-level governance) and it earned the second-best engagement rate at 4.7%. But it reached only 768 and earned zero saves, which is the exact inversion of what a framework should do. Two contributing factors. First, it shipped Aug 12, two days after the Aug 10 "MCP Evolution" post, another dense integration-layer teaching piece, so the feed saw two heavy technical posts in 48 hours and the second was throttled. Second, the three mechanisms were delivered as flowing prose deep in the body rather than as a scannable, numbered list a reader could screenshot and keep. Same substance, packaged for the bookmark, is a save magnet.
1. Thin saves: only 6 saves across 9 posts, and the frameworks (the natural bookmark format) earned just 2. Payoffs were not packaged as numbered, keepable takeaways.
2. Deep-technical clustering: two dense integration-layer posts within 48 hours (Aug 10 and Aug 12) throttled the second to the month's lowest reach.
3. Hedged authority: the usage-logs post undercut its own strong insight with repeated caveats, softening the hook and capping reach.
4. Logistics-only event teases: the "where we'll be" tease (912) reached less than half the thesis-driven recap (1.8K); event posts need an argument, not just a schedule.
Six moves, each grounded in something we observed in August.
Hot takes were 3 of 9 posts, averaged 1,531 impressions (the best repeatable pattern), and drove a third of all engagement. The winners name concrete failures (QuickBooks dual entities, Jamf missing fields). Ship 3 to 4 sharp POV posts in September, each opening on a contrarian claim about the integration or tools layer and backing it with a specific, named enterprise failure. Candidate targets: generic marketplace MCPs, DIY connector scripts, model-swapping as a cost fix, "read-only" agent pilots that never reach write.
Frameworks reached lowest (avg 800) and earned only 2 saves, despite being the format a technical audience keeps. The substance is already there; the packaging is not. Ship 2 to 3 frameworks in September as numbered, scannable, screenshot-ready lists with the payoff promised in the first two lines: "The 3 mechanisms that actually cut agent token spend", "4 questions to vet an enterprise MCP before you deploy it", "The integration-layer checklist for going from read to write". Lead with the list, not the narrative.
The TYME and Enboarder partner post was the single widest-reach post of the month (2,634) and earned 2 saves. Ecosystem and customer-win news travels because it names recognizable companies and signals momentum. Keep it to about one per month so it stays a signal rather than noise, and always anchor it in named partners or a concrete customer outcome.
The Glean:GO recap (1.8K, 5.5% engagement, the month's best rate) far outperformed the earlier logistics tease (912) because it carried an argument: the read-vs-write gap for AI agents. For September events, lead with the thesis and use the event as proof, and always follow an event with a recap that states what you learned or argued on stage.
The Aug 10 "MCP Evolution" and Aug 12 "Token spend" posts were both dense integration-layer teaching pieces two days apart, and the second was throttled to the month's lowest reach (768) despite a strong 4.7% engagement rate. Alternate the calendar: a hot take, then a lighter proof or event post, then a framework. Target a steady 2 to 3 posts a week with clear air between the heaviest technical pieces.
The usage-logs post had a superb insight ("token usage measures activity, tool calls measure work") but buried it under hedges ("unproven", "needs more work") and a stat dump, and it reached lowest of the month. When posting an experiment or a data story, open on the single quotable conclusion, keep the caveats to one line, and give the reader one thing to take away. Confidence in the hook is what earns both reach and saves.