A8gent
The log

Engineering essays on production agents.

Long-form pieces, field notes from incidents, benchmarks. Written when something is worth saying. Updated when we're wrong.

Subscribe

Get the log by email. About once a month.

  • All 160
  • Technical 43
  • Industry 41
  • Business 26
  • Comparison 16
  • How-To 12
  • Agency 9

All posts

What you will find

Field notes from automations that had to survive a real business

Every post here comes out of a build we shipped or a question an owner asked on a consult call. That is the entire editorial rule, and it is why the posts are longer and more specific than most: if something is published here, we could not find it answered honestly anywhere else.

Expect tutorials with the actual configuration left in, write-ups where the numbers are hedged rather than rounded up, and the occasional contrarian post when the common advice is wrong. We would rather publish one useful piece a month than ten thin ones a week, and nothing here exists purely to rank.

How to read the archive

  1. 1

    New to this? Start with the fundamentals, not the deep dive

    The learning guides explain what an agent is, how it fails, and what to ship first. They rarely change. The blog assumes you already have that footing and goes straight at one specific problem, which is frustrating reading if the vocabulary is still new.

  2. 2

    Building something? Search by workflow, not by tool

    The posts that help most are the ones matching the job you are automating, not the platform you happen to have a login for. Missed calls, quote follow-up, appointment reminders, and invoice chasing show up across very different businesses with almost identical mechanics.

  3. 3

    Evaluating a quote? Read the cost and hiring pieces

    If you are trying to work out whether a proposal in front of you is fair, the decision pages are more useful than any tutorial. They cover what the work actually costs to build and run, and what separates a solo consultant from an agency in practice.

A good fit if
  • Owners and operators who want the reasoning, not just the steps.
  • Builders comparing their approach against one that shipped.
Probably not for you if
  • You want a weekly AI news roundup. There are better places for that.
  • You want vendor-neutral listicles. We name what we would actually use.

Where to go next

Reading is cheap and building is not. If a post describes a problem you recognise, the fastest next step is a free automation audit rather than a weekend of trial and error.

What this collection is for

Around a hundred long-form posts, all of which came out of a build we shipped or a question an owner asked on a consult call. Tutorials with the configuration left in, cost write-ups with the numbers hedged rather than rounded up, and the occasional contrarian post where the common advice is wrong. The archive is deep enough that browsing it is a bad strategy - arrive with a question.

It is not a weekly AI news roundup and it is not a set of vendor-neutral listicles; we name what we would actually use. If the vocabulary is still new, this is the wrong hub and learn is the right one.

How to choose

Search the archive by what you are trying to do, in roughly this order of usefulness.

  1. By workflow, not by tool. Missed calls, quote follow-up, appointment reminders and invoice chasing show up in wildly different businesses with almost identical mechanics. Find the post matching the job, not the platform you happen to have a login for - and if a full build is what you want, the build guides are the finished version of the same workflows.
  2. By trade, if your industry is unusual. There are posts written for law firms, clinics, restaurants, agencies, ecommerce and recruiting, and automation by industry collects that thinking into one page per trade with the compliance notes attached.
  3. By cost, if you are evaluating a quote. The running-cost and pricing posts are more useful than any tutorial when a proposal is in front of you. Read them beside what AI automation costs, automation build cost and how to hire a consultant.
  4. By engineering topic, if you are the builder. The deep end covers evals, observability, prompt injection, retrieval and self-hosting economics. Those posts assume a codebase and pair with the framework comparisons.
  5. By what you are selling, if this is your business. There is a cluster on pricing, selling and running an automation practice, which pairs with the agency starter kit and the pricing calculator.

If you would rather be handed a shortlist, these five carry most of the weight. Start with when not to use an AI agent, which will save more readers money than anything else here. Then what AI agents can do reliably for the honest capability line, the monthly running cost breakdown for the bill nobody budgets for, the mistakes businesses make for the failure patterns, and agents versus virtual assistants if the real alternative you are weighing is a person. Builders should add the evals and testing guide and prompt injection defence.

Where this sits

The blog is the deep, specific, occasionally opinionated layer. Learn holds the fundamentals that do not change. Guides are complete builds for a small number of workflows. Playbooks describe job shapes without naming a stack, while platforms, models and compare handle the shopping decisions - what you build on, what does the reasoning, and which of two to pick.

What to do first

Do not read the archive front to back. Name the workflow that eats the most of your week, find the two or three posts that touch it, and read those. Then size it with the ROI calculator before you build anything. If a post describes a problem you recognise and you would rather skip the trial and error, the free automation audit returns a written roadmap and the service page covers what building it for you involves.

What you will find

Field notes from shipping agents, not SEO filler

Every post comes out of a real build or a real question a client asked, which means the posts are longer and more specific than most. If it is here, it answers something we could not find answered well anywhere else.

Expect tutorials with the actual config, case studies with the numbers left in, and the occasional contrarian take when the common advice is wrong. We would rather publish one honest post a month than ten thin ones a week.

New here? Start with the learning guides for the fundamentals, then come back for the deep dives.

FAQ

Questions people ask

Still stuck? Ask us directly and we will answer.

How often do you publish?

When we have something worth saying, usually a few times a month. We favor depth over frequency.

Can I suggest a topic?

Yes. The questions readers send are where most of these posts come from. Use the contact page.

Do you use AI to write these?

The thinking, builds, and numbers are ours. If a post covers a build, we actually shipped it.

Where should a beginner start?

Read the learning guides first for the fundamentals, then pick the blog post closest to what you are building.

How big is the archive?

Around a hundred posts. It is deep enough that browsing it is a bad strategy - arrive with a specific workflow or question in mind and search for that.

How is the blog different from the build guides?

A build guide is one complete workflow with the trigger, the guards, the message copy and the compliance step, written so you can follow it end to end. A blog post is narrower or more opinionated, and may not be a full build.