|
J&DENIS
CONSULTING
|
#001 · 08.26.2026
|
The last 48 hours, distilled
SIGNAL
AI
Two days of AI,
without the noise.
46 links reviewed. Four signals kept.
One tutorial to try before the next edition.
Get SIGNAL AI for free →
|
|
The useful watch, once a week
Get the next signals
An independent digest of the ethical, legal and practical topics in AI, with a directly actionable tutorial. Free, readable in five minutes.
|
|
Hello,
Over the past two days, funding rounds, robots and new models took up most of the space. But the most useful shift is elsewhere: agents are beginning to remember, to chain more actions together, and to quietly move the place of the human.
On the menu: why "keeping a human in the loop" is no longer enough, how to keep accountability verifiable, and a very simple register to frame delegation to AI.
Enjoy,
Jérôme
|
|
|
|
If you only read one minute
The four signals
MEMORY
Claude now links chat memory and Cowork. Less repetition, but a new need: knowing what is being retained.
See the source →
|
CONTROL
A human in the loop does not guarantee real oversight. Automation can also dull the very judgment it requires.
See the study →
|
ACCOUNTABILITY
You can delegate production, not acceptance. Provenance, verification and a responsible human must stay visible.
See the study →
|
INFRASTRUCTURE
The battle is shifting to inference. Latency and work-per-watt are becoming strategic metrics.
See the results →
|
|
|
01 — Usage & data
Memory makes the agent more useful — and harder to forget
Anthropic now links the memory used in Claude conversations to Cowork's. An idea prepared in chat can therefore be picked up when it is time to produce a document or take action, without starting the whole briefing over.
The real signal is not just convenience. Memory becomes a shared layer between thinking and doing. Anthropic lets you read, edit or delete stored items and excludes several sensitive categories by default. But for an organisation, the product setting does not replace an internal rule on what an agent may retain.
| My takeaway: before turning on persistent memory, sort information into three groups: storable, session-only, and off-limits. Then add a regular review of what has actually been kept. |
|
|
02 — Ethics & governance
"A human in the loop" can become an illusion of control
A position paper co-authored notably by Margaret Mitchell warns of a paradox: agents are gaining autonomy, but their interfaces and execution speeds do not always give the human the means to exercise real judgment.
Oversight can even weaken with prolonged use of automation. Approving an output is not supervising a line of reasoning. You must preserve the context, the time, the skills and the stopping points needed to challenge the agent.
The oversight test Can the human understand, challenge and interrupt — or only click "approve"? |
|
|
03 — Accountability & infrastructure
Delegating production does not transfer responsibility
A second paper proposes a useful boundary: the ethical problem does not depend first on how much work is handed to the model, but on the quality of verification and on the existence of a responsible human who owns the final acceptance.
The authors propose an epistemic audit: a structured trail of what was delegated, where contributions came from, which checks were run and who is accountable. This logic matters all the more as infrastructure accelerates: OpenAI reports more work-per-watt and lower latency for its Jalapeño chip in early tests.
Bottom line: the faster agents respond and chain steps, the more the organisation must deliberately slow down at the points where an error becomes costly or irreversible.
|
|
04 — On the radar
Three briefs, no more
1 | In medicine, uncertainty should become an output, not a hidden detail. The FLARE framework aims to explicitly build uncertainty into the factual adoption of clinical AI. Read |
2 | The web is starting to be indexed for agents, not just for humans. Keenable is betting on structured search for systems that must act and verify, rather than simply show ten links. Read |
3 | Agent performance is also becoming an energy question. In its early tests, OpenAI measures Jalapeño by latency and work-per-watt — a sign that cost of use is overtaking the race for sheer size. Read |
|
|
05 — The hands-on tutorial
Build an AI delegation register in 20 minutes
Goal: make visible what the AI produces, what the human verifies, and who owns the final decision. A single table is enough; no code required.
| 01 | Name one precise decision or deliverable.
Example: qualify a lead, summarise a file, or prepare a recommendation. Avoid vague missions like "manage the project". |
| 02 | List what is actually delegated.
Research, extraction, comparison, drafting or sending: one line per operation, with the data and tools it can access. |
| 03 | Keep the provenance of every result.
Source, date, passage or file used. A claim that cannot be traced must stay marked "to verify". |
| 04 | Define the expected human check.
Review, cross-source, independent calculation, sample test: "verify" must become an observable action. |
| 05 | Name the owner and the stopping points.
Who accepts the result? Which irreversible actions — sending, payment, rejection, publication — require their explicit approval? |
| 06 | Test five cases, two of them deliberately ambiguous.
Record the errors, disagreements and missing information. The register should learn from these exceptions before any wider rollout. |
|
Need to adapt it to your case?
Tell me the delegated task, the tool used and the decision that worries you. I'll help you sharpen the checks — or spot what should not be automated yet.
|
|
|
06 — The through-line
Autonomy without a trace is not delegation under control.
Across these two days of news, persistent memory, faster agents and fragile oversight all tell the same story: value comes from what you can delegate; trust comes from what you can still explain, verify and stop.
|
|
If this sounds like something on your plate
I help SMEs frame their AI usage, secure their data, and turn their documents into genuinely usable processes. A simple conversation about your case is enough to start.
|
|
|
Talk soon,
Jérôme Denis
JDENIS Consulting — AI advisory & applications
|
|
© 2026 JDENIS Consulting · Toulouse, France
|
|