Most Firms Still Rarely Use AI. Email Reaches Everyone.

The median company is not running a fleet of autonomous agents. It is experimenting quietly while vendors shout about transformation. OECD’s firm-level work on the adoption of artificial intelligence in firms keeps showing concentration among larger, digitalized employers, which means most employees still experience AI as rumor, not infrastructure. That is not a moral failure; it is a distribution problem.

Why “buy more licenses” fails the people you actually employ

McKinsey’s recurring State of AI research keeps returning to the same lesson: ROI follows workflow redesign, not model shopping. If your AI strategy assumes everyone will happily open another console, you are betting against how office work already flows. Harvard Business Review’s guidance on digital exhaustion is the human counterweight—every new surface competes for the same tired brains.

NIST’s AI Risk Management Framework gives adults language for trustworthy deployment, but vocabulary does not change behavior. Anthropic’s Economic Index publications have documented uneven geographic and organizational uptake of frontier assistants, with later updates noting API workloads automating office tasks including mail-shaped analysis. Translation: the people who already benefit from AI are not randomly sampled from your org chart.

Email as boring, inclusive infrastructure

TechCrunch’s reporting on email-native agent startups is not nostalgia; it is pragmatism. SMTP is the protocol your lawyers, plant managers, and franchise owners already understand. They do not need another login to get value from a model—they need a specialist on the other side of a forward.

That is the via.email pitch without the varnish. via.email is an email-based agent platform: each specialist lives at an address, you send the context you already have, and you read the answer in-thread. Tier-dependent features like attachments or web search are spelled out on https://www.via.email/pricing. What does not change is the interaction model—mail in, mail out, human edits in between.

Three agents that meet workers where OECD says friction lives

You will not onboard the whole company to a prompt-engineering club. You can onboard them to three addresses that map to weekly pain:

• Distill to Three distill.to.three@via.email turns a long thread or PDF into three bullets a manager can act on in a stand-up.
• Extract Action Items extract.action.items@via.email pulls who owes what from the same messy forwards operations already live inside.
• Extract Newsletter Insights extract.newsletter.insights@via.email digests vendor and industry email people subscribe to but never read.

Browse hundreds of public agents at https://www.via.email/agents. via.email does not read your inbox unprompted, does not send mail for you, and does not remember unrelated threads—constraints that keep expectations honest.

The comparison you should make internally

HBR’s article on how teams spend GenAI time savings asks whether minutes saved become strategy or new busywork. If your AI pilot requires a modality switch every time, you picked the wrong lever. Our earlier coverage of SMB time lost to email, context switching as a trillion-dollar leak, and when more AI tools start to hurt tells the same story from different angles: fewer surfaces beats more horsepower.

A small experiment that actually scales

Pick ten employees who are not “digital natives” by stereotype but run real decisions through email. Give them exactly three agent addresses tied to their actual job artifacts—forwarded newsletters, project threads, customer mail. Measure whether they use them without a training webinar. If yes, you have found inclusive AI. If no, the interface was never the problem; the workflow was. Either way, you learn something useful before you fund another platform tax.