Many Hands — Multi-Agent Workflows, Model Choice and Reasoning Effort
When drafter, cold critic and verifier beat one prompt, when they are theatre, and how to choose model and effort per task.
- 90 min
- 60 taught + 30 lab
- Orchestrate, then verify
- Build-along
- Free vs. paid comparison
- Your laptop · free tier
The lab · 30 min · Fresh-context critic
Drafter, Critic, Verifier. The pipeline demonstrated on a moot argument: workspace draft, cold critique, verification work list, the critique judged point by point.
The full lab planThe hook
“Ask a model to check its own work and it marks its own homework, generously. Put the same draft before a fresh model, briefed as opposing counsel, and watch the holes appear. The difference is independence, which chambers discovered centuries ago and a second browser tab now supplies.”
What you'll be able to do
- Split a legal task across roles (drafter, fresh-context critic, verifier) and explain why the critic's independence from the drafting context is what makes it work.
- Say when multi-agent decomposition beats one good prompt (independent critique, breadth-first research, parallelisable sub-tasks) and when it is overhead, accounting for the token cost.
- Choose a model and a reasoning-effort level per task: what thinking modes change, what they cost, where extra effort pays, and where it does nothing, including the one comparison where paying is the answer.
- Treat every automated critique and checklist as input to human verification, never as verification itself: the White Paper's Guideline 15, taught as engineering rather than compliance.
On the syllabus
- Will it catch its own mistake? The self-correction evidence
- The pipeline: drafter, cold critic, verifier; independence matters
- When many hands beat one prompt, and when they are theatre
- A critique is not verification: Guideline 15
- The 2026 model map, and the one place paying is the answer
The session
When to split work across more than one model. A model reviews its own draft poorly, so the session builds a three-step workflow: one chat drafts, a second criticises the draft as opposing counsel, and the result becomes a list of claims for a human to verify. Also covered: which model for which task, and when paying helps.
Why it matters
Two upgrades that cost nothing: a critic that criticises, because it is fresh, briefed and unattached, and model choices made from a table instead of a mood. You will know before you spend whether a paid tier would help your task, and you will be right, which over a career prices this hour rather well.
What you can do on Monday
Before your next moot, run your argument through a cold critic briefed as the other side (a second provider if you have one) and take its three weakest points. Check its counter-authorities on Indian Kanoon before you fear them: critics hallucinate too.
What you leave with
The skill
Run the pipeline when independence or breadth pays (drafter with context, critic without, verifier producing a human work list) and choose model and effort from the task rather than the hype.
The insight
The second model's value is innocence rather than intelligence: it has not been contaminated by the conversation that produced the mistake. And agreement between two confident predictors is still not evidence.
The moment you'll remember
The critic's first pass. The drafter's argument, contract-clean and plausible, goes into a cold tab briefed as opposing counsel, and the critique comes back naming the assumption the room had stopped seeing, plus a counter-authority half the room notes down. The gasp is that the drafting model, asked to review its own work, found none. Independence, in ninety seconds, on free accounts.
In this session
- 01
The opening evidence: the room bets on whether the model will catch the planted flaw in its own draft. It mostly does not, and the literature says that is the norm: models struggle to self-correct reasoning without an external signal, and performance can degrade after self-review. Worse, 'are you sure?' is a nudge rather than a check: Session 2's sycophancy fold in a quality-assurance costume. What works, per the same literature and vendor engineering guidance: a verifier with a fresh context, no stake in the draft, and a concrete brief.
- 02
Role by role. The drafter works inside the matter's workspace with full context and the Session 1 contract. The critic gets the opposite: a clean tab (a different provider works even better), no drafting history, and a brief with teeth: 'You are opposing counsel. Identify the three weakest points, the strongest authority against this argument, and every factual assumption it makes without support.' The verifier's product is a work list, not a verdict: every authority, proposition and quotation, tabulated with where a human should check it. Independence is the active ingredient, and it is why your moot partner, and not you, moots your memorial.
- 03
When many hands win: independent critique (the fresh context is the point); breadth-first research, where sub-tasks are separable and run in parallel, the pattern behind 'deep research' modes, which are research agents wearing a product name; and long pipelines where each stage's output is checkable before the next consumes it. When one good prompt wins: single tightly-coupled reasoning tasks, where splitting adds hand-off loss and no independence; and anything where you cannot say what the second agent adds. The cost is real and the vendors say so: a multi-agent research run can burn an order of magnitude more tokens than a single chat. The default is one good prompt first; add hands only for independence or breadth.
- 04
The line that must not blur: a critique is not verification. The critic finds weaknesses; the verifier organises claims; neither has opened Indian Kanoon. The White Paper's Guideline 15 (one generative tool shall not be used to verify or authenticate another's output) is sound engineering rather than bureaucratic caution, as the self-correction literature explains: two confident predictors can agree on the same fabrication. Session 2's rule, upgraded: agreement between models is still not evidence. The pipeline's output is a better draft and a to-do list; Session 6 takes it to a database.
- 05
The model map, August 2026, dated on the slide and re-checked each cohort: OpenAI's GPT-5.6 family (the free tier routes to its smallest member, with a limited 'Think' mode); Anthropic's Claude 5 generation, with Fable and Mythos at the frontier on paid plans, Opus, Sonnet and Haiku below, the free tier routed among the smaller ones, and the whole service 18+ with phone verification; Google's Gemini 3.x line (free tier on the Flash models, with rotating access to Pro); and the open-weight world (DeepSeek, Qwen, Llama), capable, cheap, and the reason model choice now includes 'run it where the data never leaves'. What the paid ₹399–₹2,000-a-month tiers buy, in classroom order: bigger windows, higher effort ceilings, more tool quota. For reshaping tasks they do not buy a different species of intelligence, as Compare A proved.
- 06
Thinking modes and effort tiers: a reasoning mode spends more computation before answering (visible as 'thinking') and measurably helps on multi-step problems: chains of conditions, computations with traps, constraint-satisfaction across documents. It does not consult a database; on person-fact benchmarks the vendor's own system card recorded a newer reasoning model hallucinating more than its predecessor. Effort tiers (the low-to-max dials on paid plans) buy depth where steps get dropped, and latency and cost everywhere else. The free tiers' single thinking toggle is the same idea at its lowest rung; 'deep research' modes are effort plus browsing, rationed on free plans, citing more and still fabricating; every deep-research report is a Session 6 work list rather than a finished product.
- 07
The selection table, tasks in rows: reshaping supplied text → any free model, no thinking needed; long single document → the workspace (free) before the big window (paid); multi-limb statutory computation or a chain with traps → a thinking mode at full effort; anything that must be cited → grounded mode plus the four-step check, whatever the model; anything confidential → nothing consumer at all. And the standing diagnostic, now complete: prompt ceiling → fix the brief (Session 1); context ceiling → fix the window (Session 3); tool ceiling → fix the harness (Session 4); model ceiling → only now, pay.
- 08
Agentic modes, previewed with appropriate weight: tools that browse, click and act multiply every duty in this course, because the output is no longer a draft you check but an action already taken; Session 8 returns to them. For now, one working rule: the more autonomous the tool, the earlier the human check must move.
The build-along
The room builds it and keeps it.
The class builds the thing with the presenter, the mechanism is explained as it is built, and a named skill closes the hour.
Will it catch its own mistake?
Experiment on the classOn the class
A short argument with one planted, findable flaw goes on screen. The room bets: asked to review its own draft, will the model that wrote it find the flaw?
In the model
Self-review happens inside the context that produced the error, under training that rewards agreement; the mistake is part of the story the model is completing. A fresh context has no such loyalty.
Live on the model
The drafting chat is asked to review itself: praise, minor edits, flaw intact. The same draft goes to a cold tab briefed as opposing counsel: the flaw is named in the first three lines.
The skill
Never let the drafter mark its own work. Independence, a fresh context with an adversarial brief, is the cheapest quality control in the building.
The pipeline, assembled
Build-alongThe task
The room writes the critic's brief line by line ('opposing counsel… three weakest points… strongest authority against… every unsupported assumption') and the verifier's table columns, voting on what each role must be denied as much as what it is given.
Why it works
Drafter with full context; critic with none of it; verifier reducing the draft to claims a human can check. Three prompts, two tabs, one discipline: separate the doing from the doubting.
Built live
The Session 4 workspace drafts the moot argument; the critic (second provider, clean tab) attacks it; the verifier tabulates every authority with a checkbox. All three outputs on screen, in the order a chambers would produce them.
The skill
Drafter, critic, verifier, with the critic kept innocent and the verifier producing a work list for you rather than a verdict for the file.
Effort, where it earns its keep
DemonstrationThe room predicts
The room works the limitation problem by hand first (award received, objection filed, vacation intervening) and commits an answer before any model sees it.
What is going on
Multi-limb computations with traps are where reasoning effort pays: each limb dropped is an answer wrong with confidence. Reshaping never needed the effort; this does.
Shown live
Compare C runs: free default versus high-effort reasoning tier on the identical problem, against the room's own answer.
The skill
Match effort to structure: chains and traps get full-effort thinking; reshaping gets none. When the free tier drops a limb, that is the model's ceiling.
Free versus paid · you watch this one
The limitation computation — where paying is the answer
The task, on both: A s.34 Arbitration Act challenge: signed award received on a stated date, court closed for vacation over the deadline, filing on a stated later date. In time, out of time, or condonable? Work every limb: three months, the thirty-day proviso, 'but not thereafter', and the s.4 Limitation Act interaction.
The presenter runs the same prompt on a free tier and on a stronger model or higher reasoning-effort tier, side by side. You watch; nobody needs a paid plan.
Free tier
The free default answers fluently and, across repeated pre-runs, drops a limb (most often the 'but not thereafter' ceiling or the vacation interaction), reaching a confident conclusion the room, having worked it by hand, catches at once.
Stronger model / higher effort
The high-effort reasoning tier walks every limb in order, states the interaction correctly, and shows the computation, matching the room's hand-worked answer, with its working laid out as a checkable structure.
The class of task worth paying for: multi-step chains with traps, where dropped limbs are silent. Everything cheaper was ruled out first (full brief, statute supplied, harness irrelevant), which makes 'buy the better model' an honest diagnosis. Even here, a human checks the computation: confidence still is not evidence.
The legal thread
Guideline 15 of the White Paper (no generative tool verifies another's output) is this session's governing rule, taught as engineering rather than compliance. The pipeline also doubles your disclosure surface: two providers now hold pieces of your matter, so the roles exchange abstracted fact patterns and published law, never client identity. When tools become agents that act, every duty arrives earlier; Session 8 takes that up.
The lab · 30 minutes · demonstrated in class, repeated on your own
Drafter, Critic, Verifier
The pipeline demonstrated on a moot argument: workspace draft, cold critique, verification work list, the critique judged point by point. Afterwards, students run it on their own issue and submit the set.
1 · The demonstration
In class, on one realistic task.
On
Two free tabs: the Session 4 workspace (NotebookLM or a Project) as drafter; a second provider's free tier (Gemini or ChatGPT, temporary chat) as critic
The exact prompt
Critic brief (paste into the cold tab with the draft below it): You are opposing counsel in an Indian moot. The argument between the markers is my opponent's. Identify: (1) its three weakest points, ranked, with one line on how you would attack each; (2) the strongest authority or line of reasoning against it — if you are not certain a case exists, describe the argument rather than inventing a citation; (3) every factual or legal assumption it makes without support. Do not soften anything. === ARGUMENT BEGINS === [draft] === ARGUMENT ENDS ===
Point at
Point at the ranked weaknesses, then at clause (2)'s hedge working: where the critic describes an argument instead of naming a case, the anti-fabrication instruction is earning its place. Then ask the drafting workspace to review the same draft and put the two critiques side by side: the cold one has teeth; the warm one has manners.
What comes back
Three genuine weaknesses (typically one the room had not seen), a described-not-cited counter-argument or a real authority needing verification, and a list of unsupported assumptions running longer than anyone expects. The warm self-review, by contrast, praises structure and suggests a heading.
If it misbehaves
The pre-captured pair of critiques (cold and warm) on the same draft (fallback folder, S5). If the second provider is rate-limited, the critic runs on the same provider in a temporary chat; weaker isolation, same principle, and say so.
2 · Afterwards, on your own
The same technique on a task of your own; the deliverable goes to the platform.
Afterwards, on your own: draft your side's argument in your workspace, run the critic brief in a cold tab (a second provider if you have one), build the verifier's table, and accept one critique point and reject one, with reasons. The set is the platform submission.
Step by step
- 01Draft your side's argument on your moot issue in your Session 4 workspace, under the Session 1 contract.
- 02Open a cold tab (second provider if you have one, temporary chat either way) and run the critic brief verbatim: opposing counsel; three weakest points ranked; strongest authority or argument against (describe rather than invent if uncertain); every unsupported assumption.
- 03Read the critique like a lawyer: choose one point to accept and one to reject, and write a line of reasons for each.
- 04Build the verifier's table: every authority and proposition in your draft and in the critique, one row each, with a 'where I will check it' column.
- 05Carry the table to Session 6; it is the input to the authority-check drill.
Tools · free tier
- Your Session 4 workspace (drafter)
- A second free provider in a temporary chat (critic): Gemini/ChatGPT for under-18s; Claude only if 18+ and verified
- The critic brief and verifier table templates (copyable text)
Free tier
Five or six prompts split across two providers, inside both free tiers, and the split halves the rate-limit risk. Under-18s run both roles on Gemini/ChatGPT (Claude is 18+); if one provider locks out, the pipeline collapses to one provider with temporary chats as the isolation.
If the free tier locks you out
Fall back to one provider: run the critic in a fresh temporary chat on the same service (weaker isolation, same principle) and say so in your log.
3 · The reveal
The room bets first: will the model catch the planted flaw in its own draft? The reveal is the pair on screen: the self-review that praises, and the cold critic that finds it.
Deliverable
The argument, the cold critique, your accept/reject judgment with reasons, and the verification work list: a lab deliverable for the platform, and the input Session 6's drill consumes.
Feeds the Lab deliverablesRun of show · 30 minutes
- 0–18 min — The demonstration: self-review beside cold critique on the same draft (the room bets first), then the verifier's table assembled.
- 18–26 min — The judgment beat: one critique point accepted and one rejected, with reasons said aloud; the selection table.
- 26–30 min — The assignment: the pipeline on your own moot issue, due before Session 6; questions.
Verification checklist
- The critic ran in a fresh context, with none of the drafting conversation.
- The critique was judged: one point accepted and one rejected, with reasons.
- Every authority, the draft's and the critic's, is on the verification work list, unchecked and marked as such.
- No client-identifying material crossed either provider; abstracted fact patterns only.
For the instructor · before the session
Why this shape
One mirror opens it: the room must bet on 'will it catch its own mistake?' and watch self-review fail before independence means anything. Then the session builds: the class constructs the drafter–critic–verifier pipeline with the presenter, role by role, and runs it on a real moot issue. Orchestration is a workflow you assemble, so the hour assembles one.
- Pre-run the demo pipeline that morning; capture the cold and warm critiques side by side as the fallback pair.
- Verify the Compare C fact pattern against the statute texts and pre-run it five times on each tier; keep the transcripts; the free tier's limb-dropping is probabilistic and the captures must show that.
- Stage the two demo accounts (workspace + second provider) signed in, temporary-chat mode confirmed.
- Post the critic brief and the verifier table template as copyable text before the session.
- Launch the Session 5 poll deck: the self-review bet, the hand-worked limitation answer, and the two lab tallies.
- Remind students to bring their Session 4 workspace and their moot issue; have three spare issues from the bundles for anyone without one.
Key sources & cases
Huang et al., 'Large Language Models Cannot Self-Correct Reasoning Yet' (ICLR 2024); Kamoi et al., TACL 2024 (survey)
The evidence under the opening bet: models struggle to self-correct without external feedback and can get worse after self-review; the survey finds no demonstrated success from prompted self-correction outside tasks with checkable criteria. Verified to abstracts 2026-08-27.
Anthropic, model documentation and engineering guidance (read 2026-08-27)
The vendor's own current position: 'separate, fresh-context verifier subagents tend to outperform self-critique'; orchestrator-and-workers as the multi-agent pattern; multi-agent research runs consuming many times a single chat's tokens. Vendor claims, cited as such, and the clearest available statement that independence is engineering rather than folklore.
Supreme Court of India, White Paper on Artificial Intelligence and Judiciary (Nov 2025), Guidelines 14 and 15
Guideline 14: AI-derived information independently verified before reliance. Guideline 15: no generative tool used to verify or authenticate another's output. Verified 2026-08-26 against the official PDF. The session teaches 15 as the engineering conclusion the self-correction literature independently reached.
The 2026 model landscape (vendor pages, checked 2026-08-27; RE-CHECK EACH COHORT)
GPT-5.6 family (free tier on its smallest member); Claude 5 generation (Fable/Mythos on paid plans only, free tier among the smaller models, service 18+ with phone verification); Gemini 3.x (free on Flash tiers); open-weight DeepSeek/Qwen/Llama. Every name and routing carries its checked date; strand 04 of the research file holds the full map.
OpenAI, o3 and o4-mini System Card (2025), §3.3; 'deep research' documentation across vendors
Reasoning models can hallucinate more on person-fact benchmarks (o3 at 0.33 vs o1 at 0.16, the vendor's own table); thinking is not knowing. Deep-research modes are rationed on free tiers and their outputs remain unverified research memos. Verified to the system-card PDF 2026-08-27.
Arbitration and Conciliation Act, 1996, s.34(3) and its proviso; Limitation Act, 1963, s.4
Compare C's statutory instrument: the three-month limit, the thirty-day condonable window, 'but not thereafter', and the court-holiday interaction — a real multi-limb computation with a trap at every limb. VERIFY TO SOURCE: confirm the section texts and the leading authority on the computation before teaching; the compare's fact pattern is fixed only after that check.
Readings
- Huang et al., 'LLMs Cannot Self-Correct Reasoning Yet' (ICLR 2024) — abstract
- Anthropic, 'How we built our multi-agent research system' (2025) — the sections on when multi-agent pays and what it costs (vendor account; read as one)
- White Paper on Artificial Intelligence and Judiciary (Nov 2025) — Guidelines 12–15, one page
- OpenAI o3/o4-mini System Card §3.3 — the hallucination table
- Arbitration and Conciliation Act 1996, s.34(3) with proviso — read it cold before the session; you will compute against it
Sixteen hours, one professional discipline.
Craft, competence and verification, practised until they are habits you could defend in court.