Blog
Writing on customer support, AI agents, and the open web.
All posts
- Product
The time tax has a private collector, and Atlas pays it for you
Annie Lowrey named it the time tax: the hours, money and dignity people lose getting what they're already owed. Much of it is collected by the companies we pay. Here's how Atlas takes on its three parts: learning, compliance and psychological costs.
October 17, 2026 · 8 min
- EngineeringProduct
From Utterance to Outcome: Pragmatic Competence in Task-Oriented Voice Agents
Deployed voice agents fail at pragmatics more than at parsing. A six-part architecture — interpretation frame, layered competence, persistent goals, action-keyed clarification, legible tools, and outcome-anchored evaluation — grounded in a live pharmacy deployment.
September 17, 2026 · 27 min
- EngineeringIndustry
Modeling the Puppeteer, Not the Puppet: Theory of Mind for Agent-to-Agent Systems
Why theory of mind is the wrong foundation for agent-to-agent trust — and the right residual. A framework for what to infer, what to verify, and what to bound.
July 11, 2026 · 28 min
- Industry
The great agentic acqui-hire: what six months of AI deals reveal
The agentic-AI story of the last year was told in models and demos. The more revealing story is in the deal flow — the largest tech companies absorbing agent startups' people and code, often without buying the company outright. A look at where the value is consolidating.
July 8, 2026 · 3 min
- Industry
The consumer agent arrives: bring-your-own-agent goes from thesis to default
For two years, 'customers will show up with their own AI' was a prediction you put on a roadmap slide. Over the last few weeks it started reading like an observation. What changes for brands when the counterparty is software.
July 1, 2026 · 3 min
- IndustryProduct
Agentic CX grew up: from deflection to delegation
A short while ago, 'agentic customer experience' still mostly meant a smarter chatbot. Over the last few weeks the category has visibly matured — three shifts, from deflection to resolution to delegation, and from chat-only to voice-first.
June 27, 2026 · 3 min
- IndustryProduct
Redesign the workflow, or the agent does nothing
The productivity paradox is repeating. Companies are bolting agents onto workflows built for humans and wondering where the ROI went. The gains only show up when the workflow is redesigned around what agents are actually good at — bounded, auditable, first-turn action.
June 25, 2026 · 4 min
- Industry
The comfort curve: consumers are learning to let agents act
The barrier to agentic AI was never capability — it was consent. Over the last stretch, the consumer posture has quietly shifted from 'answer my question' to 'go do it,' and that shift is climbing a predictable ladder of trust.
June 22, 2026 · 4 min
- Industry
Principal-less: the philosophy of delegated agency
What does it mean for software to act on your behalf — and what happens when your agent transacts with another agent and no human is anywhere in the loop? A philosophical look at delegated agency, the missing principal, and trust as the substitute for supervision.
June 20, 2026 · 4 min
- EngineeringIndustry
The agentic stack, 2026: what solidified and what's still soft
A year ago the agentic stack was a pile of demos. Today it's an emerging set of standards. A technical read on what has solidified — protocols, memory, tool-use — and where the real work still is: evaluation, orchestration, and the trust layer.
June 16, 2026 · 4 min
- Industry
The regulatory perimeter is moving: 2026 trends in agentic AI law
The first wave of AI law was written for models. The next wave is being written for agents. A forward look at where agentic-AI regulation is heading — identity, contract formation, liability, and disclosure — and what operators should build now.
June 10, 2026 · 4 min
- IndustryProduct
The agentic turn in customer service
Customer service is the first mass market where agentic AI has to be more than fluent — it has to act. Why service is the proving ground for autonomous agents, and what the shift from deflection to resolution demands of the brands on the other end.
June 4, 2026 · 4 min
- Industry
The two-sided future of customer service
The next battle in customer experience isn't human versus bot. It's the consumer's agent versus the brand's agent — and the layer in between is the whole opportunity.
May 16, 2026 · 4 min
- IndustryProduct
Bring your own agent: what brands do when customers show up with their own AI
Soon your customer won't open your chat window themselves — their agent will. The brands that prepare for the agent-on-the-other-side win. The ones that try to block it lose.
May 15, 2026 · 4 min
- Engineering
Know Your Agent: identity becomes infrastructure
Banking has KYC. Agent commerce will need KYA — a way to verify that an agent is who it claims to be and authorized to do what it's doing. It barely exists yet.
May 14, 2026 · 4 min
- Industry
Who's to blame when an agent gets it wrong?
An agent commits to something the person would never have approved. Four parties could be at fault, and most discussions name only two. The missing parties are why agent adoption stalls.
May 13, 2026 · 4 min
- Industry
The per-seat pricing page is a relic
Per-seat pricing was built for software humans log into. Agents don't have seats — and the pricing page is the last part of the SaaS playbook to notice.
May 12, 2026 · 4 min
- Engineering
The protocol shake-out: what to actually build on in 2026
MCP, A2A, ACP, ANP, and a half-dozen commerce protocols. You don't have to bet on a winner — you have to understand which layer each one occupies.
May 11, 2026 · 4 min
- ProductIndustry
Deflection is a vanity metric
"Tickets deflected" measures how many customers you stopped from reaching you. It tells you nothing about whether their problem got solved — and it's quietly steering CX teams the wrong way.
May 9, 2026 · 3 min
- Industry
Agent-washing is here: a field guide to spotting it
Every CX vendor now says "agentic." Most of them shipped a chatbot with a new label. Five questions that tell the difference before you sign.
May 8, 2026 · 4 min
- Industry
"Press 1 for a human": the slow death of the phone tree
The IVR maze was never built to help you — it was built to deflect you. Voice agents and verified call trees finally invert the design.
May 7, 2026 · 3 min
- Product
Your subscriptions are a part-time job
The average household runs a dozen-plus vendor relationships by hand — renewals, disputes, downgrades, hold music. That's not life admin. That's unpaid labor an agent should absorb.
May 6, 2026 · 3 min
- Industry
From chaos to coordination: what Spirit's collapse teaches us about agentic CX
Spirit's shutdown left thousands of travelers stranded. Agent-to-agent coordination is the only way customer experience survives the next one.
May 3, 2026 · 2 min
- Product
The customer support layer of the web
Why we built Teleperson on top of the page the customer is already on, instead of asking them to switch apps.
April 12, 2026 · 6 min
- Engineering
Single-step authority: how agent-to-agent should work
Our safety model for the chat co-pilot, every send is a separate user-confirmed action.
March 28, 2026 · 8 min
- ProductIndustry
Data flywheels without the data grab
Building network-effect data quality without selling user behavior.
March 4, 2026 · 5 min