Field Notes
Long-form essays from
the deployment frontier.
What we're learning as we ship agents into production departments — across healthcare, government, research, marketing, executive search, and B2B sales. No hot takes. No vendor talking points.
One essay per fortnight · no spam · unsubscribe anytime
Deploy
or it didn't
happen.
or it didn't
happen.
Featured · 18 min read
The agentic gap is not a capability gap. It's a deployment gap.
Eighteen months into the agentic-AI era, frontier models can already do most of the operational work knowledge-workers do. So why is almost nothing in production? A field report from six engagements — and a hard look at the consulting industry that's selling decks instead of deployment.
EU AI Act
in the IDE
in the IDE
GovernanceApr 14, 2026
Annex III risk-tier mapping for production agent fleets
A practical method for mapping every agent in a production fleet to its EU AI Act risk tier — before the first prompt is written. Includes the decision tree we use across engagements.
14 min read · Jakub Bareš→
Skills
over prompts
over prompts
CraftApr 8, 2026
Why we write SKILL.md files instead of long system prompts
A skill is a teachable unit. A system prompt is a configuration string. The difference compounds: one is shippable, versionable, testable; the other is a thousand-line wall of text. A walkthrough with examples.
11 min read · J. Bareš→
SaaS
was the
warm-up
was the
warm-up
EconomyApr 2, 2026
SaaS was the warm-up. Agent fleets are the main act.
Why the SaaS pricing model — per-seat, per-month — is structurally incompatible with the economics of agentic AI. And what comes next: outcome-based, share-of-value, the return of the partnership model.
16 min read · J. Bareš→
The 7-day
diagnostic
diagnostic
DeploymentMar 26, 2026
The 7-day diagnostic: how we frame an agent engagement
A close look at the first week of an Orchestrary engagement: the diagnostic interview, the workflow decomposition, the opportunity scoring, and the live Ignite demo on day six. With the actual artefacts, redacted.
13 min read · J. Bareš→
On-prem,
by default
by default
CraftMar 19, 2026
When to choose OpenClaw over Claude Code: a sovereignty checklist
Five questions we ask in the first hour of every engagement to decide whether the agent fleet should run on managed infrastructure or fully on-premise. Plus the trade-offs neither vendor wants to advertise.
9 min read · J. Bareš→
Audit log
or it didn't
happen
or it didn't
happen
GovernanceMar 11, 2026
Audit logs for agent fleets: the schema we ship in every engagement
A reference schema for cryptographically chained audit logs of agent actions — replayable, tamper-evident, GDPR-compatible. The same one we deploy on day one of every engagement, open-sourced.
12 min read · J. Bareš→
The Czech
mid-market
opportunity
mid-market
opportunity
EconomyMar 4, 2026
Why the Czech mid-market is the most under-served agent opportunity in the EU
Czech mid-market companies are technically capable, GDPR-aware, and starved of practical AI delivery. A market sizing — and an argument for why the Anglo-American AI consultancies are about to miss it.
15 min read · J. Bareš→
Eval gates
in CI
in CI
DeploymentFeb 25, 2026
Golden datasets and eval gates: stopping a regression before it ships
A walk-through of the agent evaluation pipeline we wire into every client's CI before the first cutover. With code examples in Python and the GitHub Actions YAML we actually use in production.
17 min read · J. Bareš→
Small
tools, big
leverage
tools, big
leverage
CraftFeb 17, 2026
The single most overlooked skill in agent engineering: writing small tools well
The difference between a chatbot and an agent that ships work is a handful of small, reliable, idempotent tools. A taxonomy of the tool patterns we reach for again and again — with anti-patterns.
10 min read · J. Bareš→
topics we keep coming back to
Four ongoing investigations.
If you want to follow one specific thread, these are the four we update most.
Deployment
Field reports from real engagements: architecture, integrations, cutover, on-call. The unglamorous work of getting agents into production.
12 essays · updated weekly
Governance & AI Act
EU AI Act risk tiers, GDPR, audit logs, kill switches, conformity packages. Practical compliance for working engineers.
8 essays · updated fortnightly
The agent economy
Pricing, business models, the Czech & Central European mid-market, the death of per-seat SaaS. Long-form essays only.
6 essays · updated monthly
Craft & tooling
Skills, tools, prompts, evaluation. The day-to-day texture of writing reliable agentic code in 2026.
14 essays · updated weekly
stay close to the work
Get one essay every fortnight.
Direct from the IDE.
Every Field Notes essay is written by an Orchestrary partner who's currently shipping agents into a client's production stack. No syndicated content, no guest posts, no SEO filler.
or email us at hello@orchestrary.com