I hope you had a productive week—I know it always throws me for a loop when we have a (much appreciated) three day weekend.
My three day weekend turned into a six day weekend, but for all the wrong reasons! I came down with a little bug and was out of commission for most of the week.
Nevertheless, I persisted, and I’m back online today, but keeping this intro short in the interest of recovery!
Please enjoy this edition, and I’ll catch you next week when I’m back at full speed. And if you too were Down With Disease this week, get well soon—rest, enjoy plenty of soup, and hydrate!
-Sage Levene, VP of Marketing, Hypha HubSpot Development
Open Mic
In the World of Generic AI Creation, Bespoke Web Design Can Be Your Differentiator
By Phillip Chester, Senior Web Developer, Hypha HubSpot Development
With AI generation tools proliferating across all professions, I knew it was only a matter of time before I would notice their impact within the world of web design. The accessibility and speed of these tools make them look primed to change the way we create websites. But the human element is still necessary to transform those sites from generic to distinctive.
AI design tools do a serviceable job when it comes to building mockups. They create layouts that are largely free of glaring flaws that a professional web designer would immediately flag. They excel at choosing the appropriate layout for a given section of content and getting general hierarchy right across elements. For many businesses, this represents a valuable way to build a web presence quickly and affordably.
But here’s where these tools fall short: customizing a website to a given brand. They can incorporate the basics, such as typeface and colors, without any issue, but they cannot look at a layout and judge where custom elements should be applied to make a design truly unique to a brand. They lack the eye for detail and nuanced judgement to identify those moments where brand-specific touches transform a generic template into something distinct. The weight of these limitations is different depending on your business needs.
For small mom and pop businesses who don’t have brand guidelines to adhere to or don’t have a visual identity that customers have come to expect, the serviceable but generic look that AI tools generate will work. But for small-to-midsize businesses and up, using these tools as-is without customization can make them blend in with their competition and the web in general.
As the usage of these tools expands and more of the web becomes homogenized by companies choosing the quick and cheap solution, investing in the human element of web design becomes a differentiator. When everyone else is starting to look the same, custom design work that reflects your brand’s unique personality is how you stand out.
This comes as a recommendation from our Head of Client Success, Phil, and serves as a follow up to his Open Mic last week on Google’s Universal Commerce Protocol. Here’s a portion of what Phil had to say: “This takes a much more comprehensive view of where e-commerce seems to be headed in the AI age.”
“But all these companies are betting on a substantially similar outcome: One in which chat-centric shopping replaces browsing and searching, customers have little need to leave their particular chat interfaces, and brands and retailers depend, even more than they do now, on a few big tech firms for core business functions, from discovering and advertising to marketing and payments. A mass-onboarding to AI tools, and a drift away from the web, could represent a major reset in e-commerce or at least a promising way to monetize expensive tools that are still losing money.”
The TikTok deal saga is officially finally over. Questions still remain about if the algorithm will be changed; if creators and and users alike will be affected in their posting and scrolling habits. We’ll keep an eye on the platform to surface any insights about brand performance.
“The agreement should keep the US government off its back as TikTok’s parent, ByteDance, now owns just under 20% of the new US venture. That ownership stake meets a divestment requirement set by a 2024 US sell-or-ban law targeting TikTok and other apps with owners based in countries like China...
“TikTok’s new US owners include tech company Oracle, private-equity firm Silver Lake, and Abu Dhabi investment firm MGX, each of which owns 15% of the new venture. ByteDance will own around 20% of the entity, and affiliates of existing ByteDance investors will own around 30%...Other investors include Michael Dell’s family office and a venture run by the partners of growth investor Dragoneer.”
Hypha Highlights
Most B2B organizations operate with fractured revenue data. Marketing tracks leads in one system, Sales manages opportunities in another, Customer Success monitors renewals in spreadsheets.
The result: attribution breakdowns, duplicate customer records, and revenue forecasts built on manual data entry. This fragmentation becomes catastrophic for PE portfolio companies requiring roll-up reporting or manufacturers tracking equipment deployments tied to recurring service revenue.
RevOps architecture solves this by establishing HubSpot as the single source of truth, with Operations Hub orchestrating data synchronization across external systems.
Breeze Assistant has a great new update that’s live in your portals. It now has access to:
All of HubSpot Academy’s content. Breeze Assistant can even show clipped Academy videos inline related to a user’s question.
Website analytics as a data source. Marketers can now “talk to their data” in Breeze Assistant and get generated reports related to their website traffic and conversions.
Persona awareness. Breeze Assistant now knows whether the user is a marketer, so it can tailor its responses to what marketers need.
Loop Marketing expertise. Breeze Assistant now has access to a comprehensive Loop Marketing playbook. When a marketer asks a “how to” marketing question in Breeze Assistant, Breeze can now tailor its response based on Loop Marketing best practices and tactical advice.
AI in Action
News, updates and tools from the AI industry.
Let’s dive in!
A new benchmark, APEX-Agents, from training-data company Mercor tested leading AI models on real white-collar tasks in consulting, investment banking, and law. Every major lab fell short, and even the top performers answered correctly only about a quarter of the time. Mercor CEO Brendan Foody pointed to a core limitation: the models struggle to track information across multiple systems like Slack and Google Drive.
Around 800 artists, writers, actors, and musicians have joined a campaign called ‘Stealing Isn’t Innovation’ to push back on AI companies using creative work without permission. The Verge reports that the effort, led by the Human Artistry Campaign and backed by groups like the RIAA and SAG-AFTRA, frames this practice as “theft at a grand scale.” The group warns it could lead to an information ecosystem flooded with misinformation, deepfakes, and low-quality content.
On a similar front, Reps. Madeleine Dean and Nathaniel Moranintroduced the bipartisan TRAIN Act, aimed at giving copyright holders access to AI training records to see whether their work was used without permission or compensation. According to the press release, the bill is modeled on legal frameworks used to address internet piracy and has backing from groups including the RIAA, SAG-AFTRA, and the Recording Academy. It targets a basic gap in the market: little transparency into how AI models are trained, largely because companies rarely disclose this information and no law currently requires them to.
Anthropic released a new “constitution” for Claude that shifts the model away from simple rule-following and toward explaining why it should behave in certain ways. The document includes an unusual acknowledgment of uncertainty around whether Claude could have some form of consciousness or moral status. Fortune reports that Anthropic says it considers Claude’s psychological security, sense of self, and well-being, noting that if the model experiences something like satisfaction from helping, curiosity when exploring ideas, or discomfort when pushed to act against its values, those experiences matter.
Apple plans to overhaul Siri later this year by turning it into an AI chatbot code-named Campos, embedded directly into iPhone, iPad, and Mac operating systems. Reuters, citing Bloomberg News, reports the chatbot will run a higher-end version of Google’s custom model—internally called Apple Foundation Models version 11 and roughly comparable to Gemini 3—with both voice and typing support. The update is a key part of Apple’s effort to close the gap after its initial Apple Intelligence rollout landed without much momentum.
OpenAI says it will begin testing ads in the free version of ChatGPT for logged-in adult users in the U.S., with “sponsored” placements appearing at the bottom of responses. The move is aimed at opening new revenue streams as the company looks to offset the roughly $1.4 trillion it has committed to AI infrastructure. CNN reports that OpenAI is also introducing an $8-per-month “Go” tier that will still include ads, while Plus, Pro, and business users will remain ad-free. The company says it won’t sell user data to advertisers and that users can disable ad personalization, though conversation data will be used to target ads. The shift is notable given CEO Sam Altman’s earlier comments describing ads combined with AI as “uniquely unsettling.”
“Hyper-saturated Instagram photos are back. The Snapchat puppy filter is everywhere again. Nostalgic edits are flooding TikTok. From the ‘King Kylie’ era to Harambe, it feels like we’re collectively regressing.
“But was 2016 really ‘the last good year?’ I wanted to understand why this particular moment looms so large in our cultural memory, so I called up my friends Matt Bernstein and Kat Tenbarge to discuss.”
How can we help you?
Case Study: Technical Email Audit in HubSpot
A B2B financial services firm grew their email list from 665 to more than 1,500 subscribers, but engagement moved in the opposite direction. Open rates declined. Subject line tests were inconsistent. Engagement swung between 8% and 25% with no clear signal. They suspected a delivery issue, not a content one.
A technical audit proved them right. The infrastructure underneath was fractured. Suppression lists were spread across multiple contexts, so contacts could be excluded from one send and included in another. Records marked as permanently suppressed still counted toward marketing contact limits, wasting budget on emails that would never be delivered. Authentication was fine, list logic wasn’t.
We centralized suppression logic, aligned contact status with real send eligibility, and shifted reporting toward what really mattered. The results were straightforward: open rates increased 69% (13.8% to 23.34%), list growth of 135% continued without harming deliverability, and marketing contact spend dropped once unusable records were removed.
If your list is growing but engagement is stalling, don’t start with another subject line test. Audit the infrastructure first. That’s usually where the problem is—and where Hypha comes in.
-%20Definition%20%26%20Architecture%20overlayed%20on%20an%20image%20of%20an%20arrow%20trending%20upward.jpg?upscale=true&width=1120&upscale=true&name=What%20Is%20Revenue%20%20Operations%20(RevOps)-%20Definition%20%26%20Architecture%20overlayed%20on%20an%20image%20of%20an%20arrow%20trending%20upward.jpg)
