CURRENT | Foundations

Field Guide

When Web Agents Need to Remember What They're Doing

By Rina Takahashi— November 13, 2025

Feature image for article: When Web Agents Need to Remember What They're Doing

A web agent halfway through extracting hotel inventory across regional booking systems hits a rate limit. Should it remember where it was? Compare that to another agent checking a single price in three seconds. One workflow needs to preserve an hour of authentication and navigation state. The other can restart faster than you could save its progress.

Operating web agents at scale, this distinction shapes infrastructure complexity, recovery mechanisms, what you can reliably run. State persistence isn't binary. It's a spectrum, and understanding where your workflows fall determines what you actually need to build. Most teams make architectural commitments before they have that map.

Field Guide

When Web Agents Need to Remember What They're Doing

A web agent halfway through extracting hotel inventory across regional booking systems hits a rate limit. Should it remember where it was? Compare that to another agent checking a single price in three seconds. One workflow needs to preserve an hour of authentication and navigation state. The other can restart faster than you could save its progress.

Operating web agents at scale, this distinction shapes infrastructure complexity, recovery mechanisms, what you can reliably run. State persistence isn't binary. It's a spectrum, and understanding where your workflows fall determines what you actually need to build. Most teams make architectural commitments before they have that map.

Rina Takahashi

Rina Takahashi, 37, former marketplace operations engineer turned enterprise AI writer. Built and maintained web-facing automations at scale for travel and e-commerce platforms. Now writes about reliable web agents, observability, and production-grade AI infrastructure at TinyFish.

Web Archaeology

The Twenty-Four Hour Problem

By Rina Takahashi— November 13, 2025

Feature image for article: The Twenty-Four Hour Problem

During a site migration, our web agents encounter something strange: different parts of our infrastructure see different versions of reality. DNS resolvers return different IP addresses for the same domain. Requests time out. Authentication fails. A 1983 architectural choice, working exactly as designed.

The number that protected root servers on a research network now creates windows where truth is relative and time-delayed. Operating web automation at scale, we can't assume DNS resolution returns consistent answers. We build around it instead.

Web Archaeology

The Twenty-Four Hour Problem

By Rina Takahashi— November 13, 2025

During a site migration, our web agents encounter something strange: different parts of our infrastructure see different versions of reality. DNS resolvers return different IP addresses for the same domain. Requests time out. Authentication fails. A 1983 architectural choice, working exactly as designed.

The number that protected root servers on a research network now creates windows where truth is relative and time-delayed. Operating web automation at scale, we can't assume DNS resolution returns consistent answers. We build around it instead.

In Dialogue With Complexity

An Interview With 600 Mbps, the Bandwidth Threshold Where HTTP/3 Stops Working

In Dialogue With Complexity

An Interview With 600 Mbps, the Bandwidth Threshold Where HTTP/3 Stops Working

Pattern Recognition from the Field

Everyone Built Agents Before Building Agent Governance

Six vendors launched AI agent identity management in November 2025. Microsoft, AWS, CyberArk, Ping Identity, Oasis Security, Akeyless. All solving the same problem: agent sprawl.

Here's what that timing tells you. 82% of companies already run AI agents. 53% of those agents touch sensitive data daily. But the infrastructure to track them, secure them, manage their lifecycles? Just showing up now.

Watch what happens in practice. Agent ownership changes hands four times in the first year. Developer leaves, agent stays. Still holding credentials, still accessing systems. No owner, no lifecycle, no audit trail.

The industry shipped agents first, asked governance questions later. Now we're retrofitting identity management onto deployed systems. Build it in from the start. Your future security team will thank you.

Pattern Recognition from the Field

Everyone Built Agents Before Building Agent Governance

Six vendors launched AI agent identity management in November 2025. Microsoft, AWS, CyberArk, Ping Identity, Oasis Security, Akeyless. All solving the same problem: agent sprawl.

Here's what that timing tells you. 82% of companies already run AI agents. 53% of those agents touch sensitive data daily. But the infrastructure to track them, secure them, manage their lifecycles? Just showing up now.

Watch what happens in practice. Agent ownership changes hands four times in the first year. Developer leaves, agent stays. Still holding credentials, still accessing systems. No owner, no lifecycle, no audit trail.

The industry shipped agents first, asked governance questions later. Now we're retrofitting identity management onto deployed systems. Build it in from the start. Your future security team will thank you.

Deployment stall:

Pilots jumped from 37% to 65% in one quarter, but full production stuck at 11% because governance infrastructure wasn't ready.

Ownership drift:

Agents pass from executive sponsor to AI team to Cloud Ops during year one, creating gaps when original builders leave.

Standards lag:

Human and application identities are well understood. Agent identities remain unsettled, with standards catching up to what's already deployed.

Cost chaos:

Teams spin up agents independently. Result is overlaps, redundancies, runaway API calls, agents looping endlessly and overloading systems.

Protocol emergence:

Google donated Agent2Agent to Linux Foundation in June. Microsoft announced broad MCP support across platforms in May.

Questions Worth Asking

The questions you ask before choosing a tool matter more than the vendor answers you get. We've watched organizations stumble not because they picked the wrong technology, but because they asked the wrong questions. Or skipped the hard ones entirely.

Teams focus on features and demos while missing what predicts success at scale. They evaluate tools in isolation instead of asking how those tools will interact with their actual processes, data, and people.

These six questions cut through that noise. They're what we return to when evaluating anything meant for production, because they reveal what actually matters once the demo ends and real work begins.

Questions Worth Asking

The questions you ask before choosing a tool matter more than the vendor answers you get. We've watched organizations stumble not because they picked the wrong technology, but because they asked the wrong questions. Or skipped the hard ones entirely.

Teams focus on features and demos while missing what predicts success at scale. They evaluate tools in isolation instead of asking how those tools will interact with their actual processes, data, and people.

These six questions cut through that noise. They're what we return to when evaluating anything meant for production, because they reveal what actually matters once the demo ends and real work begins.

Performance Baseline

Can You Define Success Before You Launch?

Service Level Objectives aren't paperwork. They're your shared definition of acceptable performance. Without them, every incident becomes a debate about whether things are actually broken. When monitoring and ownership exist before launch, mean time to recovery drops because responders know exactly where to look and who owns what.

Compliance Scaling

Will Compliance Costs Scale With Your System?

Most organizations focus on passing the first audit. The better question: how does your architecture affect ongoing compliance effort? Systems where evidence generates automatically and violations are prevented by design cost dramatically less to maintain than those where compliance is bolted on afterward.

Process Prerequisites

Does This Assume Your Processes Already Work?

Technology amplifies what you already do. If your processes are messy, software makes them messier faster. The tools that work best enhance existing clarity rather than attempting to impose structure where none exists. Look for solutions that fit how your business actually operates, not how vendors think it should.

Data Reality

Will This Expose Your Data Problems?

Planning tools need timely, structured, accurate data. They won't fix inconsistent formats or missing records. Poor master data produces unreliable outputs regardless of how sophisticated the software is. The question isn't whether the tool can handle your data—it's whether your data is ready for the tool.

Daily Usage

Would Your Team Actually Choose This?

Powerful software fails when users reject it. Clunky interfaces and inflexible reporting reduce adoption faster than missing features. The real test isn't whether the tool can do the job. It's whether your team will want to use it every day, under pressure, when easier alternatives exist.

Integration Proof

How Many Production Integrations Actually Exist?

Look for documented APIs, plug-and-play integrations, and published examples of successful deployments with tools your organization already uses. Custom stacks with no documented API and few integration examples signal future pain. Vendor promises about integration capabilities matter less than evidence of working implementations.