The Business Analytics Institute offers a wide range of training, coaching and consulting services to help management improve their ability to take tough decisions.
A 12-Part LinkedIn Series · Prelude to BAI Winter School 2027, Cape Town
A person in the loop
A person in the loop
isn't the same as a person in control.
Twelve articles testing whether "human in the loop" oversight of agentic AI is real or symbolic — across the EU, US, UK, South Korea, and China. Named sources only. No orphaned statistics.
Subscribe to the Series Read the Docket
What This Series Is
One gap, tested across five legal systems
Every major regulatory model for high-risk AI shares one instinct: put a human somewhere in the decision chain. The series opens with The Warm Body in the Loop, naming the gap between a human being present and a human being in control — using legal scholars Sebastian Schwemer and Jenni Koivisto's distinction between formalistic and substantive oversight as its spine, then testing it against the EU's Article 14, Colorado and Texas's retreats, the UK's silence, South Korea's grace period, and China's traceability-first bet.
The arc runs in three moves: The Illusion (Articles 1–4) diagnoses why oversight fails in practice; The Reckoning (5–8) works through the architecture, the contested terminology, and the agentic systems that act before a human ever sees them; The Advantage (9–12) turns to what oversight built to actually work looks like — closing with the bridge to BAI Winter School 2027 in Cape Town.
Who reads it: compliance and risk leadership building toward the EU AI Act's delayed conformity deadlines, practitioners designing human-oversight architecture rather than waiting to be told to, and anyone deciding whether their own "human in the loop" is a safeguard or a description of furniture.
Exhibit A — Brussels
Dec 2, 2027
The Digital Omnibus pushed the EU's high-risk conformity-assessment deadline — where Article 14's oversight obligations actually bind — more than a year past the date enforcement powers went live. The regulator conceding its own verification machinery wasn't ready is itself evidence for the series' thesis.
Exhibit B — Three Agentic Gaps
Accountability · Reasoning · Authority
The BAI/Kore.ai framework, mapped to EU AI Act Articles 9, 12, and 14, runs underneath every article as shared vocabulary — reused, not redefined, article to article.
Why It's Worth Following
Six things this series won't let you assume
Presence isn't control
The formalistic/substantive distinction, applied consistently, article to article, so "human in the loop" stops functioning as a synonym for "solved."
Five jurisdictions, one map
The EU, US states, UK, South Korea, and China each built — or declined to build — the same requirement differently. The series tracks all five without flattening the differences.
Scale, not just attention
Article 2 shows a capable, attentive reviewer still hits a hard ceiling — one reviewer, a million decisions — regardless of how sharp any individual's judgment is.
The market already knows
A vendor cluster now sells "proof of genuine oversight" — behaviorally detecting hollow approvals. Nobody builds a product to detect a problem that isn't real.
Named sources, no orphaned figures
Every specific number without a checkable source is written as hedged, observational language — not asserted as settled fact. That discipline holds across all twelve articles.
Diagnosis before design
The first eight articles hold the line on naming the failure before proposing a fix — so Move 3's "advantage" answers a problem the series actually proved, not one it assumed.
The Docket
Three moves, twelve articles
Move 1 — The Illusion
Diagnosing why "a person present" keeps getting mistaken for "a person in control"
Is the oversight in front of you real, or is it furniture?
01
The Warm Body in the Loop
Formalistic vs. substantive oversight, tested against the EU, US, UK, South Korea, and China — the gap named, jurisdiction by jurisdiction.
16 Aug 2026Published
02
One Reviewer, One Million Decisions
Even a fully attentive, empowered reviewer runs into arithmetic: scale mismatch, from AML queues to code review to security operations centers.
30 Aug 2026Publishing
03
The Deference Reflex
What automation bias does to judgment over time — and why it compounds with, rather than duplicates, the scale problem above.
13 Sep 2026Coming
04
Skating Blind
Deskilling and cognitive offloading — what sustained reliance does to a reviewer's own capability to catch an error at all.
27 Sep 2026Coming
Move 2 — The Reckoning
Working through the architecture, the contested terms, and systems that act before a human sees them
If the default model is broken, what is it actually asking of the people building oversight now?
05
Article 14 and the Architecture of Attention
Inside the EU's five-capability standard — what it actually requires of a human overseer, and why the machinery to verify it isn't ready.
11 Oct 2026Coming
06
Meaningful Control, Contested Meaning
"Meaningful human control," from autonomous-weapons ethics to enterprise agentic AI — and why it isn't the same thing as Colorado's statutory "review right."
25 Oct 2026Coming
07
The Four-Eyes Problem
Why doubling up reviewers doesn't double oversight — and where the compliance-detection vendor market gets a fuller look.
8 Nov 2026Coming
08
When the Agent Acts First
The Three Agentic Gaps — Accountability, Reasoning, Authority — applied to systems where "in the loop" arrives after the decision, not before it.
22 Nov 2026Coming
Move 3 — The Advantage
Designing oversight that answers the failure modes named in Moves 1 and 2 — and the evidence it pays off
What does oversight look like when it's built to actually work, not just to be seen?
09
Designing for Dissent
A direct callback to Articles 1–4: mitigations built to visibly answer formalistic oversight, scale mismatch, deference, and deskilling in turn.
6 Dec 2026Coming
10
The Trading Floor Test
Inside BAI's Anglet capstones — fraud detection, credit decisioning, AML triage — as working oversight architecture, not theory.
20 Dec 2026Coming
11
Teaching the Overseer
What training a genuinely capable human overseer actually requires — drawn from BAI's own curriculum design.
3 Jan 2027Coming
12
The Human Stake
The series closes — and opens the door to Cape Town: "Beyond Automation: The Human Stake in Agentic Systems," BAI Winter School 2027.
17 Jan 2027Coming
Follow the Series
Receive each article on publication
Biweekly, 16 August 2026 – 17 January 2027. One email per article, direct from the author.
GDPR Compliant · Unsubscribe Anytime · No Spam