Documentation mess just happens when smart people move fast for years.
You can't out-process entropy.
The companies waiting around to "organize everything properly first" are getting lapped by teams who just accepted the mess and plugged in something like Rovo to make sense of it anyway.
Perfect systems are for the ideal world. Searchable mess is actually achievable.
I have been using Jira for over 20 years, and we added Confluence about 6 years ago. Unfortunately, I have found Rovo to be both annoying and unhelpful. Rovo's prompts on the pages frequently interrupt the flow of my work on Jira and Confluence pages, getting in the way of what I need to get done, and disrupting my concentration rather than adding value.
Getting support from Atlassian has long been a weak point, with the company relying heavily on the Atlassian Community Forums, where users bitch, share complaints, and attempt to help one another troubleshoot Atlassian errors, bugs, and changes that were not thoroughly thought out before they were implemented.
Rovo does nothing to address the support shortcoming. Instead, Rovo feels like yet another attempt by Atlassian to shirk their responsibility to properly support paying customers.
Thjs reasonated, especially “here’s the source” vs “I think we agreed on”. I currently do LLM evaluation/data QA work (dataset validation + output checks), and I’m trying to grow into AI product/ops in bigger orgs. What’s the fastest way you’ve seen newcomers build credibility, platforms to learn on, or project path you’d recommend? If you’re open, I’d really appreciate a DM.
Funny that you're talking about Rovo. Atlassian is trying to be the "data hub" for all the data in an org, not just confluence and jira (they wanted to buy Glean a couple years ago, did not work, then they built Rovo. Rovo is a wrapper that uses the semantic layer of Jira and Confluence, with automations capabilities. In organizations like where I work we have Rovo, ChatGPT enterprise … it's not trivial to know what tool to use for what, but ChatGPT is making good strides in this "I'm going to be able to read all your corporate documents" … and now they added a "Rovo connector". basically all those large vendors (MS, Atlassian) are trying to lock the customer in their own environnement and OpenAI and Anthropic are trying to become that environment where you do all your work. Interesting battle ...
agreed the only issue is that the knowledge base is usually in multiple systems and getting intelligence from the data means gathering it from those multiple, disconnected systems. Knowing what happened last week on customer XYZ means pinging Jira and Confluence, MS 365 (email, teams messages), Salesforce (deal data), Gong (conversation transcripts), maybe the ERP, the time management system, the invoicing system …
Some vendors have their "semantic layer" = MS, Atlassian, Salesforce.
None has a "universal semantic layer".
So maybe the agnostic players - like Anthropic or Glean have the best position because then need to build connectors with everything.
Teams don't have to change their habits or spend insane hours organizing folders when they have a system like this that can understand their entire history and then point them to the right decision.
It’s a massive win for productivity.
Invisible barriers slow down organizations no matter the size, but this fixes that.
Literally a central knowledge hub for the entire team.
I generally read all your reviews fairly thoroughly and I compare your comments to those of Gary Marcus who is very conservative on AI and its impact and implementation etc.
I respect both of your analysis and I tend to agree for the most part with Gary because in general technology adoption zooms up the ladder quickly to find there are definitely issues that need to be addressed at a slower pace and past has indicated as such 2000 technology boom 2008 mortgage collapse and many others probably further in the past.
This current article of yours which I have not read in great detail but seems to indicate your backing off and being more conservative because AI is not the end all be all as many hyperscaliers are professing. What worries me most is the circular involvement of the big seven or eight or even the other 10 or 12 at the second tier level.
I read a lot about the impact on the economy the sources of power source of cooling and the tie in to the global banking system. All of this technology is so intertwined such as Elon musk's starlink and global use of satellite or not Earth technologies for the future. I wonder how all this is going to play out in the next five to 10 years and would appreciate less of the day-to-day implementation of AI and maybe an article or two on your prognosis for the future of humanity in the next 5 to 10 years. I think that would be very useful for many of your readers if you would take a broader view which maybe you have but I haven't seen it and print yet.
In summary I very like your approach to AI and would like to read more less from implementation but more from a real practical and theoretical impact on not only AI in general but in humanity has a whole.
Sorry for being long-winded.
I am a physicist with a background in practical implementation of technology at Intel and related companies in the 80s 90s and 00s.
My bias is practical implementation because that’s where the signal is.
Most AI impact talk stays abstract.
The places it’s already real are boring and repeatable: finding the right decision, pulling the right doc, cutting rework, reducing time lost to internal search.
That’s why I liked Rovo. It turns messy history into answers with sources, without pretending the org is about to become autonomous.
On the macro side, the constraints you listed are the story for the next 5 to 10 years: energy, cooling, capex concentration, governance, and fragile dependencies.
I’ll write a broader piece if enough readers want it :)
I'm currently being put in charge of researching how to implement AI at my company. Can you help me with the how, useful steps and/or suggestions and guidelines to follow?
I do love using AI for research and also writing letters especially complaints lol it also helped me to gain my diploma in cybersecurity and networking
AI is at its best when it turns scattered info into something you can act on fast, and when it helps you draft with clarity (and this one does it best for me: https://ruben.substack.com/p/opus?r=5m7l8v)
I recently contracted at a large retailer who use Confluence. My role was to consolidate several years worth of Confluence spawl. Rovo is good but not perfect. I analysed Rovo’s performance pulling data from Confluence pages with and without meta data. Some stark differences.
I also built & tested a separate chatbot as a comparison with and without knowledge graphs, converting Confluence pages to markdown and then chunking. This gave me better visibility and control over the user query reasoning processes. Again, different/more relevant outcomes as compared to the Rovo responses.
Rovo’s edge is speed-to-value. It plugs into Confluence where the mess already lives and gets teams to answer plus source fast, without a months-long cleanup or rebuild.
Metadata still matters. But the practical move is: ship something people use now, then improve over time.
Documentation mess just happens when smart people move fast for years.
You can't out-process entropy.
The companies waiting around to "organize everything properly first" are getting lapped by teams who just accepted the mess and plugged in something like Rovo to make sense of it anyway.
Perfect systems are for the ideal world. Searchable mess is actually achievable.
The org will never finish organizing.
What matters is being able to ask a question and get the answer plus the original decision, doc, and context.
Rovo meets teams where they already work and turn years of messy history into something you can navigate without a clean-up project.
I have been using Jira for over 20 years, and we added Confluence about 6 years ago. Unfortunately, I have found Rovo to be both annoying and unhelpful. Rovo's prompts on the pages frequently interrupt the flow of my work on Jira and Confluence pages, getting in the way of what I need to get done, and disrupting my concentration rather than adding value.
Getting support from Atlassian has long been a weak point, with the company relying heavily on the Atlassian Community Forums, where users bitch, share complaints, and attempt to help one another troubleshoot Atlassian errors, bugs, and changes that were not thoroughly thought out before they were implemented.
Rovo does nothing to address the support shortcoming. Instead, Rovo feels like yet another attempt by Atlassian to shirk their responsibility to properly support paying customers.
Hi, Edward
The way I see it is simple: ChatGPT for what your team already knows in Jira + Confluence.
It should help you find the right page, ticket, or decision fast, answer, and show the source so you can trust it.
If the prompts are breaking your flow, that’s fixable.
Thjs reasonated, especially “here’s the source” vs “I think we agreed on”. I currently do LLM evaluation/data QA work (dataset validation + output checks), and I’m trying to grow into AI product/ops in bigger orgs. What’s the fastest way you’ve seen newcomers build credibility, platforms to learn on, or project path you’d recommend? If you’re open, I’d really appreciate a DM.
You’re already in the right lane. QA + eval is the fastest feeder role into AI product/ops :)
DM me and I’d try to help.
Most AI feels like hype until it finds the doc you needed yesterday
Found it in seconds is the real unlock.
Funny that you're talking about Rovo. Atlassian is trying to be the "data hub" for all the data in an org, not just confluence and jira (they wanted to buy Glean a couple years ago, did not work, then they built Rovo. Rovo is a wrapper that uses the semantic layer of Jira and Confluence, with automations capabilities. In organizations like where I work we have Rovo, ChatGPT enterprise … it's not trivial to know what tool to use for what, but ChatGPT is making good strides in this "I'm going to be able to read all your corporate documents" … and now they added a "Rovo connector". basically all those large vendors (MS, Atlassian) are trying to lock the customer in their own environnement and OpenAI and Anthropic are trying to become that environment where you do all your work. Interesting battle ...
Yep. It’s a battle.
But orgs need a shared workspace where the entire company context is available without anyone copy-pasting sensitive stuff into a random chat.
My take? the winners will be whoever combines two things best:
the company knowledge base (Jira + Confluence history), and
speed of adoption (saving hours every week).
Everything else is toothpaste AI.
agreed the only issue is that the knowledge base is usually in multiple systems and getting intelligence from the data means gathering it from those multiple, disconnected systems. Knowing what happened last week on customer XYZ means pinging Jira and Confluence, MS 365 (email, teams messages), Salesforce (deal data), Gong (conversation transcripts), maybe the ERP, the time management system, the invoicing system …
Some vendors have their "semantic layer" = MS, Atlassian, Salesforce.
None has a "universal semantic layer".
So maybe the agnostic players - like Anthropic or Glean have the best position because then need to build connectors with everything.
The token limit has increased 800 times, with an approximate price moving to 5.3 times, quite astonishing.
That’s why Rovo is interesting.
The win is retrieval and answers grounded in your workspace.
The copy-paste problem is the thing nobody talks about.
You can't upload client docs to ChatGPT. So you summarize. Badly.
And now you're doing extra work just to use a tool that was supposed to save you time.
That's why Rovo looks interesting.
Not the AI writing stuff. The "show me where this decision lives" part. I need it to remember where stuff is and finds the right page fast.
That's just... useful.
The win now is answer plus source.
Show me the page, the decision, the thread, the ticket.
In the tools we already use. With permissions intact.
Rovo gives that faster access to what the org already knows.
I am curious about implementing Rovo in my manufacturing company. Would love to get your help - Toriqul from Bangladesh
Hi, do you run Confluence/Jira?
Good article. Also it’s fun to see you still use Wispr. This is the sleeper tool of 2025.
I use it literally every day :) here’s my full review on it too: https://ruben.substack.com/p/flow?r=5m7l8v
Teams don't have to change their habits or spend insane hours organizing folders when they have a system like this that can understand their entire history and then point them to the right decision.
It’s a massive win for productivity.
Invisible barriers slow down organizations no matter the size, but this fixes that.
Literally a central knowledge hub for the entire team.
Teams keep working in Jira, Confluence, Slack, and Drive.
Rovo turns all that history into something you can query.
Entropy stays. The access friction goes away.
I generally read all your reviews fairly thoroughly and I compare your comments to those of Gary Marcus who is very conservative on AI and its impact and implementation etc.
I respect both of your analysis and I tend to agree for the most part with Gary because in general technology adoption zooms up the ladder quickly to find there are definitely issues that need to be addressed at a slower pace and past has indicated as such 2000 technology boom 2008 mortgage collapse and many others probably further in the past.
This current article of yours which I have not read in great detail but seems to indicate your backing off and being more conservative because AI is not the end all be all as many hyperscaliers are professing. What worries me most is the circular involvement of the big seven or eight or even the other 10 or 12 at the second tier level.
I read a lot about the impact on the economy the sources of power source of cooling and the tie in to the global banking system. All of this technology is so intertwined such as Elon musk's starlink and global use of satellite or not Earth technologies for the future. I wonder how all this is going to play out in the next five to 10 years and would appreciate less of the day-to-day implementation of AI and maybe an article or two on your prognosis for the future of humanity in the next 5 to 10 years. I think that would be very useful for many of your readers if you would take a broader view which maybe you have but I haven't seen it and print yet.
In summary I very like your approach to AI and would like to read more less from implementation but more from a real practical and theoretical impact on not only AI in general but in humanity has a whole.
Sorry for being long-winded.
I am a physicist with a background in practical implementation of technology at Intel and related companies in the 80s 90s and 00s.
Hi, Thomas
No need to apologize.
My bias is practical implementation because that’s where the signal is.
Most AI impact talk stays abstract.
The places it’s already real are boring and repeatable: finding the right decision, pulling the right doc, cutting rework, reducing time lost to internal search.
That’s why I liked Rovo. It turns messy history into answers with sources, without pretending the org is about to become autonomous.
On the macro side, the constraints you listed are the story for the next 5 to 10 years: energy, cooling, capex concentration, governance, and fragile dependencies.
I’ll write a broader piece if enough readers want it :)
I'm currently being put in charge of researching how to implement AI at my company. Can you help me with the how, useful steps and/or suggestions and guidelines to follow?
Hi, send me a DM :)
I do love using AI for research and also writing letters especially complaints lol it also helped me to gain my diploma in cybersecurity and networking
AI is at its best when it turns scattered info into something you can act on fast, and when it helps you draft with clarity (and this one does it best for me: https://ruben.substack.com/p/opus?r=5m7l8v)
I recently contracted at a large retailer who use Confluence. My role was to consolidate several years worth of Confluence spawl. Rovo is good but not perfect. I analysed Rovo’s performance pulling data from Confluence pages with and without meta data. Some stark differences.
I also built & tested a separate chatbot as a comparison with and without knowledge graphs, converting Confluence pages to markdown and then chunking. This gave me better visibility and control over the user query reasoning processes. Again, different/more relevant outcomes as compared to the Rovo responses.
Totally fair.
But most orgs never sustain the ops.
Rovo’s edge is speed-to-value. It plugs into Confluence where the mess already lives and gets teams to answer plus source fast, without a months-long cleanup or rebuild.
Metadata still matters. But the practical move is: ship something people use now, then improve over time.
Very true especially for orgs that shy away from dedicated in-house AI dev teams