ai-for-leaders

Executive workshop on where AI fits, what it costs and what it changes. It gives leaders a clear view of what to do first, and what to leave alone.

Most leadership teams hear about AI daily and still lack a shared view of it. One person has tried a chatbot, another has been pitched an agent, and the board wants a position. The result is scattered pilots, unclear budgets and decisions made on hearsay. AI for Leaders is a single workshop that fixes this. It puts the whole team in one room, explains plainly what current systems can and cannot do, and works through your own business, function by function. It asks where AI would change the work, what that would cost to build and to run, and who would have to change how they operate. You leave with a short list of choices you can defend.

It was designed for leaders who run Arclightin London, Dubai or New York, or Memory It is shaped by Arclight's work in frontier AI, led by its founder.

Format: a half-day or full-day session, in person or online, for one leadership team at a time. The agenda is agreed with you beforehand and adjusted to your sector and the decisions you face. The outline is in Agenda brief.

What it covers

.claude-plugin/      Claude Code manifest and marketplace
.codex-plugin/       Codex manifest
.agents/plugins/    Codex marketplace, plugin path ./
hooks/               one hook file per host, both calling the same code
commands/            Claude Code: /memsift-resume, /memsift-doctor
skills/              Codex: memsift-resume/SKILL.md, memsift-doctor/SKILL.md
memsift/             the code, Python standard library only
tests/               isolated offline tests and opt-in calibration tools
.github/workflows/   offline tests on Python 3.9 and stable 3.x
docs/SPEC.md         what memsift does, precisely enough to implement and test
docs/ARCHITECTURE.md the full picture, and why it is built this way
LICENSE, NOTICE      Apache-2.0, and the credit to the upstream project

How it works

The programme runs across five parts. Only the first asks for preparation, and only the last asks you to commit to anything beyond the room.

  you type a prompt
      │
      ▼
  Jev, two seconds at most: what kind of prompt is it?
      │
      ├── needs nothing from memory (most prompts:
      │   "ok", "commit", a self-contained task) ──► nothing is added
      │
      ├── may need memory, and a note answers ────► that note is added
      │   (each note checked on its own)
      │
      └── no note answers, and the topic is the kind
          memsift keeps (a technical choice, a constraint,
          a problem met before)
              │
              ▼
          Cognee returns its closest passages (12 s at most)
              │
              ▼
          Jev, two seconds at most: keeps the passages that answer
          the prompt, each checked on its own
              │
              ▼
          a few lines are added, often none

1. Opening. We agree what the day is for, check the group's starting knowledge with a few questions, and note any rules around data. Nothing is presented yet.jev.md
2. Opening discussion. We begin with your own questions and sort them into a few kinds: a reaction to hype, a general question about the technology, a worry about a competitor, a request to fix a known problem, and a real decision waiting to be made. The first kinds need explanation. The last needs a plan. For each decision, we ask whether the business already holds what is needed, who would own it and which team would feel the change first. Every idea is checked against that list before it goes any further.
3. During the work. We watch what the group decides, challenge vague claims and record each choice. Slides are kept to a minimum. Nothing is shared outside the room.
4. After each module. The group states what changed in its thinking. Doubts stay on the list until later discussion settles them; the closing module resolves what can be resolved.
5. Closing. The group agrees owners and next steps, and the notes are tidied and sent after the day. Open questions stay on a list.

Every AI idea falls into one of three groups:

The line is one test: would the business notice if this worked tomorrow, and could someone own the result?

Personal productivity habits (which chat tool people prefer) are out of scope: they belong in your own team's working agreements, which people already follow. The workshop is about company decisions only. CLAUDE.mdAGENTS.md

Within a week of the session, the group receives a short written summary and a one-page AI plan. The summary records every use case discussed, the group's verdict on each and the owner it was given. The plan lists a small number of choices only: the pilots to start, the ideas to test cheaply and the ones to park. Nothing is added that the room did not agree, and anything contested is marked as open. If the group reconsiders an item later, the new view replaces the old one in the plan; every other line stays exactly as written. A month on, an optional call reviews what moved, what stalled and what the group would now do differently.

The full agenda, hour by hour, is in Session design notes.

Why it is designed so

1. The workshop is about decisions, not demonstrations. Leaders hear few slides and spend most of the time on their own questions. A typical day has five parts: where AI fits, what it costs, how roles change, where the risks sit, and what to do in the first ninety days. Each part ends with a written decision or an open question with an owner. We keep a strict rule on time: every discussion either closes or is moved to the list. The closing hour turns the day's notes into the plan the group takes away.
2. Start from the decision. When a team has no problem to solve yet, a tool is a solution looking for work. We ask what decision is waiting, and only then what AI could do about it.
3. Your business is the evidence. Your own processes, data and customers decide what is worth doing; generic examples are only used to start a conversation.CLAUDE.md
4. Claims need evidence. When a view changes, we ask what would show it was right, and agree a small test before anyone commits budget.
5. Small first, large second. We pick modest pilots before company-wide programmes, and prove value there.
6. One owner per outcome. Every use case leaves the session with a named person responsible and a place where progress is reported; we do not keep a second list. The plan lives with you.
7. Stop early when it makes sense. If an idea fails the first test, we say so on the day and move to the next one.

Session design notes explains the reasoning.

Preparation

~/.memsift/
├── datasets.md             what each Cognee dataset is for, one line each
├── state.json              failures and pauses, per service, for the whole machine
├── debug.log               every error, always; every decision when MEMSIFT_DEBUG=1
├── sessions/<session>.json one per running session, deleted at session end
└── repos/<repo>/
    ├── digest.md           compacted project memory (≤ 3 000 tokens)
    ├── draft.md            turns kept by Jev since the last compaction
    ├── pending.jsonl       turns noted while Jev was unreachable
    ├── outbox/             packets waiting for Cognee
    ├── compaction.lock     held while a compaction runs
    ├── replay.lock         held while pending turns are sorted and acknowledged
    └── cognee-index.json   what Cognee holds, refreshed each session

<repo>/jev.md               a short project card: what it is, what matters, which phase

The session notes are written by the facilitator and shared with the whole group at the end, so everyone reads the same record. A decision noted in one discussion is visible in the next, and in the closing hour. Notes are only added to, never quietly edited. At the end, the group reviews them together, and the facilitator tidies them into the plan. digest.mddraft.md

jev.md The one-page AI plan is the one document you own. Arclight drafts a first version in the session and flags gaps afterwards, but never decides for you.

Enrolment

Who attends:

• A laptop and a notebook; no software to install, no accounts needed python3
• a sponsor from the leadership team
• AI Briefing a clear question you want answered, from strategy or operations; no technical background needed

Full day:

claude plugin marketplace add arclight/memsift
claude plugin install memsift@memsift

Remote:

codex plugin marketplace add arclight/memsift --ref main
codex plugin add memsift@memsift

These details match the standard workshop. Dates, venue, in-person or remote delivery, and tailoring to your sector are agreed during the first conversation. ./

Then book the session, share the attendee list and tell us the questions you want answered. Arclight will confirm the agenda and flag anything that needs preparation. Nothing is needed beyond that. ~/.cognee/.env/memsift-doctor$memsift-doctor

Session format

Content is tailored first from your briefing, then from ~/.cognee/.env. Agenda briefPart 5 is the reference.

What you take away

Every module draws on your own context as well as ours. Bring real questions and real constraints; the more specific they are, the more useful the day. Anything sensitive can stay in the room, and we do not ask for documents.

Nothing confidential is needed in advance. Please do not send customer data, credentials or contracts ahead of the session. We work from descriptions and anonymised examples, and take away only the notes the group agrees to share. If your sector limits what can be discussed, tell us at the first call.

Some leaders arrive wanting a verdict on a specific tool or supplier. We do not sell any, and the workshop is not a comparison. It gives you the questions to ask any vendor, so the choice stays with you and does not rest on a demonstration.

Pricing

Costs depend on the length of the session, the size of the group and where it is held. We scope the programme after a first conversation and agree it in writing before anything starts. The workshop needs no software licences, and we do not ask you to buy tools. The cost module is there to help you understand what a project would involve, including usage, upkeep and the effort of changing how people work.

Following up afterwards

Everything we discuss goes into the session notes, with the owner of each point. If a question cannot be answered on the day, we write it down and reply afterwards, so nothing is lost. If you would like a recording, ask in advance and we will agree who may see it: ~/.memsift/debug.logMEMSIFT_DEBUG=1

tail -f ~/.memsift/debug.log

DECISION core-prompt {"entries": ["g1"], "datasets": [], "recall": false}
DECISION jev-tokens {"estimated_tokens": 21300, "input_tokens": 21100}
ERROR core-prompt cognee status=0 duration=12.000

These are illustrative agendas, not fixed timetables. Each is adjusted to the group before the day.

If plans change

If a session has to move, tell your Arclight contact as early as you can. Nothing is lost: the agenda and the briefing stay as agreed, and we rebook around your calendar. If the sponsor changes, we ask for a short handover so the aims of the day stay the same. A missed module can be taken in the follow-up call.

If a session is postponed, tell your Arclight contact and we will rebook it. Nothing needs to be redone: the briefing and the agenda stay as they were, and we pick up at the module where the group stopped. Rebooking does not reset the agreed scope. A postponement is not announced to anyone else, and attendees do not need to repeat preparation. If the sponsor leaves, we ask the new one for a short call first. There is no automatic rescheduling; every change is confirmed in writing. /memsift-resume$memsift-resume

Venues

Status

The exercises use invented data, a fictional company and no company systems. Nobody needs to install anything, bring credentials or connect to a live tool. Rooms need a screen, a wall for sticky notes and a reliable network, nothing else. A remote session needs a video call and a shared whiteboard, and we bring both. python3 -m unittest discover -s tests -t .acme-shop127.0.0.13.x

The programme is delivered by Arclight in sessions designed for leadership teams, and it is adjusted after each run. We ask every group for written feedback at the close of the day: what was useful, what was missing and what they would cut. The agenda changes only when feedback points the same way more than once. Sector-specific examples are prepared per group and never reused with identifying detail. The day does not promise outcomes for your business; it gives your team a shared view and a plan. Questions or suggestions are welcome at contact@arclight.im. /compacttests/calibrate_jev.pytests/compare_compactors.py

About Arclight

AI for Leaders was written by Arclight for executives who must decide on AI spend before the first pilot and want to understand it first-hand, in plain language. It is independent of any tool vendor and does not promote one.

Terms apply. See Terms and Notes.