The Next Competitive Advantage Is AI Connected to Everything

If you don't want to read the unreadable blog, then please watch this Unwatchable Episode 4

TL;DR

The businesses pulling ahead with AI right now are not the ones with the flashiest model. They are the ones who have connected AI to every system they run: CRM, timesheets, invoicing, email, and call transcripts. Computan's leadership team calls this a "hyperconnected office," and in a recent conversation, CRO Joe Jerome and CEO Sajeel Qureshi walked through real examples: an AI-built deal pipeline that outperformed HubSpot, a case study generator that turns old projects into new proposals in under an hour, a plain-English HubSpot module builder, and a podcast ROI audit that uncovered hidden revenue. They also unpacked why Claude Fable is quietly handing off a large share of its coding work to Claude Opus 4.8, and what that means for developers. The throughline: connected, context-rich AI beats a smarter model working in isolation.

Why "Connected" Beats "Smart"

Most conversations about AI still center on which model is best. Computan's team thinks that misses the point. In their view, the real advantage comes from how deeply AI is wired into the rest of the business.

Joe Jerome, CRO, Computan put it plainly: "Really where we see the market going is it's gonna go into these setups where you have an AI connected office and you're running your own skills and your own agentic flows more on the custom side rather than a fixed or canned SaaS kind of solution."

Sajeel Qureshi, CEO, Computan framed the same idea around a familiar pop culture reference: "I'm looking forward to the day where everybody has a Jarvis. They walk into a room and they just talk to him or her and say, hey, what do I gotta do today. That could be everything from a work meeting to you gotta take your kid to a dentist appointment. It'd have everything in there for you."

That kind of always-on assistant does not come from a single chat window. It comes from feeding AI a constant stream of company context: recorded meetings, transcripts, emails, CRM records, and invoicing data. Computan calls their internal version of this "Cerebro," a nod to the Professor X device that gives access to everything happening across an organization at once.

The "Company Brain" in Action

The most convincing part of the conversation was a string of specific, low-drama examples of AI doing real revenue and operations work once it had full context.

1. A deal pipeline built without touching HubSpot

Joe described feeding Cerebro every sales transcript, email, and call recording and asking it to reconstruct the sales pipeline from scratch. "We had Claude go through all this stuff and create a deal pipeline for us. Give us the value of the deal, the next step, who owns it, who's responsible, what we should be on the lookout for. Literally this whole breakdown, and it went through with several deals, did not miss one. Its accuracy was way better than HubSpot."

2. Case studies built from old tickets in under an hour

When a white label partner needed proof of an SAP integration for a client pitch, Joe turned to Claude instead of digging through old project files by hand. "I gave it the scenario, I gave it the context it needed... it went through and formatted them, Computan branded, made them beautiful. This whole process took me 30 to 45 minutes, and it was immensely helpful to what we were working on."

Sajeel had a similar experience while preparing a bid. He needed to confirm whether Computan had prior experience integrating specific NetSuite entities into HubSpot. "I went to Claude and said, can you tell me if we've ever done integrations with NetSuite involving these three or four unique entities. It went through all the tickets, it went through all the NetSuite integrations that we did, and it uncovered that yes, we had done work with two of those entities."

3. A plain-English HubSpot module builder

Computan's developers built an internal tool that turns a written description into a working HubSpot module. Joe explained: "It's a HubSpot module creator that connects right into the HubSpot page." Sajeel added the mechanics: "You connect your portal, you type, give me a module, give me a testimonial widget with headshot, logo, and text box, and boom, there's your module, just like that."

Their point was not that this replaces developers. For anything sophisticated, Joe was clear a developer still needs to be involved. The value is in how fast a starting point now comes together.

4. Turning a podcast into a revenue attribution report

Computan runs a companion podcast called RevOps 500. Joe wanted to know whether it was actually driving business. He pulled the list of everyone who had booked a recording slot through HubSpot and asked Claude to cross-reference it against invoicing and email history. "It labeled them as a chicken or an egg. Eight out of ten were chickens," meaning they were already customers before appearing on the show. "But it did give us two eggs, and those two eggs were significant amounts of business."

That analysis also surfaced a gap: roughly seventy to eighty guests had never been properly followed up with after their episode. Claude scored each one by title and website fit and drafted a starting point for outreach, which Joe still edits by hand before sending.

Claude Chat vs. Claude Cowork: Why the Tool Matters

Part of what made the deal pipeline project possible was using the right tool for a multi-step job. Joe distinguished between quick chat-based prompts and longer sequential work: "You really can't hand off that agentic work to Claude Chat like you can Claude Cowork." The first pass, pulling raw deal data from transcripts, ran in Claude Chat. The second pass, evaluating and prioritizing each deal, ran as a longer sequence in Claude Cowork, taking about 45 minutes end to end.

Why Fable Is Quietly Handing Off Work to Opus

The conversation also touched on something timelier: reports of Claude Fable underperforming on coding benchmarks since its return. Joe had looked into a benchmarking report and found the headline numbers were misleading, because they counted model refusals as automatic failures.

The real story, according to Joe, is that "the new Fable is now handing off its tasks to Opus. We're finding it's like 45 to 70% of the time on coding. So when you're using Fable now, 45 to 70% of the time in these tasks, it is going back to Opus 4.8 instead of giving you Fable 5."

Sajeel summed up the dynamic in a single line: "So it's subcontracting the job to Opus."

Joe's read on why this is happening ties back to safety controls: "This was all done because they determined that Fable was so good at coding that it was a security threat... it's got a very high false positive rate on what could be malicious intent. Anything remotely close to an edge case in refactoring, debugging, or UI kind of coding, it says, nope, this might be a security risk, go to Opus."

The two of them also discussed what this means for developers. Sajeel noted the mixed reaction: "The average developer or the weak developer's probably happy because now there's gonna be some sort of a short demand spike. But the good developers are gonna be kicking themselves because now they have to do more on their own."

Joe connected it to a broader pattern in how AI has actually changed development work: "The senior developer versus the junior developer, that's where AI's really come in. It's helped people with what would be junior developer, maybe even intermediate level developer work. But when it comes to the people who are really good at this, it's aiding them, not taking their expertise away."

Where the Real Opportunity Sits: The Two Million to Twenty Million Dollar Gap

A recurring theme was who benefits most from this connected approach. Joe pointed to a specific segment: businesses doing between two million and twenty million dollars a year in revenue, who are too big to run on spreadsheets but too small to justify an enterprise HubSpot subscription or a dedicated HubSpot admin.

"There's a real opportunity to empower folks in that range, going direct with Claude," Joe said. "Sure, you can use the HubSpot CRM to store data, but it's not going to be your front end."

He also offered a word of caution about how far to push AI into customer-facing work. On a recent website refresh project, Joe was direct about the limits: "I don't suggest anyone go and mail in website work with Claude. I've seen a lot of good sites that were good sites before that have turned into really bad sites, in my opinion, because people have deferred a hundred percent to Claude. But with the appropriate workflow, the appropriate context, and the appropriate design inspiration made by humans, you can come up with a pretty good workflow."

The Bigger Shift: Bespoke Over Canned

Joe tied all of this back to a broader prediction about where AI tooling is headed. Off-the-shelf agentic features, he argued, are running into real limits: "This is why AI is being dismissed, because it's not useful. We can't get in our operation. I even talked to some very smart people in the SaaS world who are building tools with AI and they've had to change their game plan, and they're having trouble deploying products where it's useful. Ultimately the reason they're having struggles with that is because Claude pretty much does most of the stuff with its connectors, and a lot of the bespoke homegrown stuff is having the most value."

That is the core argument behind the "AI connected to everything" idea. A generic AI feature bolted onto a SaaS product will always be limited by what that product can see. A company that connects its own AI instance to its own timesheets, CRM, invoicing, and communications has something no vendor can sell off the shelf.

FAQs

What does it mean for AI to be "connected to everything" in a business?

It means an AI system has direct access to a company's core data sources, such as its CRM, timesheet and ticketing system, invoicing, email, and meeting transcripts, rather than being limited to a single chat window with no memory of the business. This is done through custom connectors, often built as MCP (Model Context Protocol) integrations, and through tools like Claude Cowork that can run longer, sequential, multi-step tasks.

Why is Claude Fable handing off coding tasks to Opus?

According to Computan's internal benchmarking, Fable is routing a significant share of coding requests, an estimated 45 to 70 percent, to Claude Opus 4.8 rather than completing them directly. Joe Jerome's assessment is that this is a safety measure: because Fable proved highly capable at tasks like penetration testing and code injection, it now flags a wide range of edge cases in refactoring, debugging, and UI work as potential security risks and defers to Opus instead.

Does AI replace developers?

Not according to Computan's leadership. Their view is that AI has mainly absorbed junior and intermediate level development work. For senior developers, it acts as an aid rather than a replacement, since it still cannot reliably handle the most complex edge cases without human judgment.

Which businesses benefit most from a connected AI setup?

Computan points to companies generating between roughly two million and twenty million dollars a year in revenue. These businesses are typically too large to manage growth manually but too small to afford an enterprise HubSpot subscription or a dedicated HubSpot administrator, making a direct, AI-connected workflow a practical middle path.

Should AI be used to build a company website end to end?

Computan's team advises against fully deferring website work to AI. Their recommendation is to use AI within a defined workflow that still includes human-provided context and human-made design direction, rather than letting AI make design and content decisions unsupervised.

What is the difference between Claude Chat and Claude Cowork for business workflows?

Claude Chat works well for single, quick requests. Claude Cowork is built for longer, sequential, multi-step work, such as evaluating dozens of sales deals in one run. Computan uses both together: Claude Chat for a first pass at pulling raw data, and Claude Cowork for the longer analysis and prioritization that follows.