Welcome back to The Hypha Wire, and happy first summer Friday for those who celebrate.
Last week, we shared a sentiment from Hypha CEO Jed’s piece about where HubSpot is headed with their agentic vision, and it got a lot of traction. Good timing, because this week there’s more to add to that conversation.
HubSpot dropped the CLI announcement (more on that below), and between that and the Agentic Customer Platform rollout, it’s worth reiterating about what’s happening: HubSpot is making a serious architectural bet. The platform isn’t just adding AI features on top of what already exists. It’s reframing the entire product around the idea that AI agents need structured customer context to do anything useful, and that the CRM has to be the place where that context lives.
Jon Sasala wrote a deep breakdown of this on our blog, and the part that sticks with me is the framing around data quality. Agents aren’t going to fail dramatically when your CRM data is messy. They’re going to produce confident, plausible, wrong answers, and no one will immediately know. That’s a different kind of problem than a broken workflow.
The CLI announcement fits into this same thread. The use cases HubSpot describes—scheduled pipeline scans, automated account reviews, recurring enrichment gap reports—are only as good as what’s in the CRM when the agent runs them. Clean data isn’t a nice-to-have in that model, it’s the whole premise.
The teams that get the most out of where HubSpot is going probably aren’t the ones who adopt the fastest. They’re the ones who have done the foundational work. That’s where we’re focused.
-Sage Levene, VP of Marketing, Hypha HubSpot Development
Open Mic
Stop Buying Features. Start Designing Workflows.
By Amy Dodge, HubSpot Platform Specialist, Hypha HubSpot Development
Most HubSpot portals don’t fail because the software is too complicated. They fail because they were built around features, not around how people actually sell, support, and market day to day.
One test that cuts through the noise: Can a rep open their portal and know, within 10 seconds, what to touch next and why it matters? If they can’t, the system is serving the software, not the business.
Almost every audit surfaces the same three problems: shadow systems (spreadsheets, personal inboxes, side Slack threads—workarounds for a CRM nobody fully trusts), unspoken exceptions that quietly break every automation, and no single owner of the full customer journey. Sales owns one slice, marketing owns another, ops owns a third.
The fix isn’t more tools. It’s drawing one honest picture of how work actually moves, then bending HubSpot to match that, not the other way around.
Clean the data model before you touch automation. Automation on top of messy data doesn’t save time; it just lets you send more confusion, faster. Decide which objects matter, which properties drive decisions, and what makes a record good enough to act on.
Then design for your least technical user. Build the 9 a.m. view before you build the dashboard: one saved view, a clear daily routine, work from the top, log what happened, trigger the next step right there.
That’s the difference between a team that’s been “trained on HubSpot” and a team whose tools finally match the way the business actually runs.
Settle in for a nice, long read about the history of machine-generated text. Hint: it’s not a new practice.
“Well, then you’d have to wonder: Who wrote this, and why, and how...For Knapp and Michaels, meaning without intention does not exist: ‘What a text means and what its author intends it to mean are identical.’ An author without an intention, they argued, is not an author.”
“This week, an AI training startup called Shift said it would clean New Yorkers’ homes for free. But there’s a catch. There’s always a catch.
“In exchange for the cleaning, Shift wants footage of its cleaners at work: scrubbing dishes, wiping counters, dusting tables, mopping floors. It wants everything. Video of all the boring domestic labor we’d happily outsource if we could — and that robotics companies are racing to teach machines to do so they can sell us something to do it for us.”
Hypha Highlights
Evaluating a HubSpot Solutions Partner means verifying that a specific agency can handle the specific complexity of your implementation—not confirming they hold a tier badge.
A complete evaluation covers five dimensions: partner tier and what it actually measures, accreditations and the requirements behind them, discovery call behavior, technical and integration depth, and team delivery structure.
The failure mode this checklist protects against is predictable: an agency wins the sales process on presentation quality and fails on delivery, leaving behind a portal configured around HubSpot features rather than the company’s actual business processes. Use these questions before signing on any engagement involving data migration, custom integrations, or architectural decisions that are difficult to reverse.
“The HubSpot Agent CLI brings HubSpot’s data and intelligence into the environments where GTM and ops teams are composing their own workflows – Codex, Claude Cowork, and Claude Code – and allows agents to automate repetitive, bulk, and scheduled work.
“It’s built on the same foundation as our public APIs and MCP server that already power our AI Connectors — and it’s designed to complement them, not replace them. AI Connectors are great for work where a human is in the loop, talking to an agent: questions, insights, conversations, campaign analytics. The Agent CLI can help agents accomplish those tasks, too, but it’s particularly useful for the repetitive, bulk, and scheduled work that needs to run without a human in the loop.”
AI in Action
News, updates and tools from the AI industry.
DeepSeek is nearing a $10 billion funding round at a roughly $45 billion valuation, with backing expected from China’s national AI fund, Tencent, and several other investors, as founder Liang Wenfeng has told potential backers the company will prioritize AGI research over commercialization. The fundraising comes alongside the company’s decision to permanently slash prices on its flagship V4 Pro model, keeping output tokens at $0.87 per million—roughly 34 times cheaper than GPT-5.5—in what analysts are calling an increasingly direct price war with Western AI labs.
As AI costs spiral, Google is positioning its Gemini 3.5 Flash as a cheaper alternative to frontier models, with CEO Sundar Pichai claiming top cloud customers could save over $1 billion annually by shifting workloads to Flash—a strategy backed by Google’s ownership of the full AI stack, from custom TPU chips to data centers, which analysts estimate cuts its internal compute costs by 50-75% compared to rivals.
Pope Leo XIV released his first papal encyclical, “Magnifica Humanitas,” a sweeping 42,300-word document calling for government regulation of AI companies, protections for workers displaced by the technology, and strict ethical constraints on AI-enabled weapons—presented alongside Anthropic co-founder Christopher Olah.
OpenAI CEO Sam Altman said he no longer believes AI will cause a “jobs apocalypse,” walking back years of bold predictions about widespread job displacement and saying he was “delighted to be wrong,” citing the resilience of human connection in the workplace as a key factor he had underestimated. His about-face comes as some observers note the timing aligns with OpenAI’s push toward an IPO and growing public skepticism about AI.
Anthropic released Opus 4.8 just 41 days after Opus 4.7, an unusually fast upgrade cycle that comes amid lukewarm reception to the previous release, with the new model notable for being more likely to flag uncertainties and avoid unsupported claims. The release also includes a new feature called Dynamic Workflows, designed to help manage complex tasks across hundreds of parallel subagents.
Additionally, Anthropic has raised $65 billion in new financing at a $900 billion pre-money valuation, officially surpassing OpenAI’s last valuation of $730 billion to become the world’s most valuable AI startup, with a revenue run rate that crossed $47 billion this month.
“Connecting with Google CEO Sundar Pichai at I/O every year is one of my favorite Decoder traditions. This was our fifth year doing it, and there’s always a whole slew of new things to talk about. This year, in addition to the news, we talked about Google Zero; picking fights with YouTube creators and publishers; and what being at ‘the foothills of the singularity’ even means.”
How can we help you?
Case Study: HubSpot Chatbot + Knowledge Base
A hyperbaric medicine provider had spent years producing educational content to rank for clinical searches. When HubSpot’s AI chatbot tools became viable, the knowledge base problem that typically stalls these projects was already solved. Most teams start from scratch. This client didn’t.
We restructured existing SEO blog content into HubSpot knowledge base articles, then built a hybrid system: AI chatbot handles open-ended clinical questions drawing from the knowledge base, traditional chatflow handles structured requests—appointments, direct contact, care team handoffs. Neither component does the other’s job.
The result: visitors get substantive answers outside business hours from content the provider has been refining for years. New content produced for search now expands the chatbot simultaneously—content calendar and knowledge base are no longer separate concerns.
Already sitting on years of blog content and considering an AI chatbot? Contact Hypha—what you’ve already built may be closer to chatbot-ready than you think.
