『656. Jean-Christophe Lanoix, Turboconsultant: Running a Solo Practice on AI Agents, end-to-end』のカバーアート

656. Jean-Christophe Lanoix, Turboconsultant: Running a Solo Practice on AI Agents, end-to-end

656. Jean-Christophe Lanoix, Turboconsultant: Running a Solo Practice on AI Agents, end-to-end

無料で聴く

ポッドキャストの詳細を見る

プレミアムプランが3か月 月額99円

登録で500円分のAmazonギフトカードプレゼント。12月23日まで。※適用条件あり
概要
Jean-Christophe Lanoix is the founder of Turboconsultant. He spent seventeen years at Hinicio, a strategy consultancy specialising in hydrogen — joining as an intern, rising to Associate Director and leaving in 2024, two years after the firm's exit. He now builds the system he wishes he had had. Introducing Turboconsultant An execution layer for solo consultants: a virtual team of AI agents spanning the whole practice — business development, research, methodologies, marketing and content, deliverable production, quality control, meetings and follow-up, admin. The consultant directs. The agents execute. It installs as a plugin on Claude Cowork and turns a generalist agentic workspace into a consulting-grade one. Three things separate it from a general-purpose assistant. Personalisation. A one-time onboarding hands it the consultant's own methodologies, templates, frameworks and voice. What comes out arrives in their template, follows their methods, and reads in their words. Quality control. Every document it produces passes three layers of audit before it leaves — sources traced, figures recomputed, claims contested by models outside the system. That is where AI hallucination and error get caught. Data walling. Nothing moves from one engagement to another, and the models do not train on client work. And it is one system rather than a patchwork of tools: the whole practice runs in a single environment that sharpens with every engagement. The net effect, measured on the existing user base, is a 25x on average, while improving the general quality of the work. Demonstration of Turboconsutlant Both cases are invented, and so is the consultant behind them — a generic profile carrying no proprietary methodologies, no past deliverables, no writing to learn from. The engagement starts the way a real one does: a request for proposal lands by email. Everything that follows comes out of that one document and public sources. The proposal — thirty minutes The system interviews the consultant first — how they want to work, whether they know the client, who delivers, what their day rate is. Then it writes, in their own template: context, the question behind the client's question, objectives, approach and methodology, scope in and out, work packages, data room requirements, timeline, budget options, next steps. The consultant reviews and signs off. Nothing leaves unvalidated. Without AI, a proposal of that depth is several man-days. The methodology section — six work packages mapped onto one governing question. Pricing the work — three options, fees, payment terms. A commercial proposal, not an analysis. The kick-off pack — fifteen minutes One instruction produces the whole pack: the client kick-off deck, an internal briefing note — assumptions, questions to ask, objections to expect, the landmines — and the data room request list in Excel, item by item, with what each one unblocks and the date it is needed by. In a firm, this is junior work. A solo consultant has no junior. It costs them half a day to a full day, every time. The work-package sequence agreed at kick-off, and the date after which the analysis stops waiting for client data. The First Deliverable — forty-five minutes The pattern holds whatever the sector: the research, the market sizing, the model underneath, and a deck of exhibits in the consultant's own consulting-grade template — each exhibit carrying one argument, each figure carrying its source. This is where the consultant sits as a director rather than an executant: give the instruction, let the team execute, come back and review. It is also where the system refuses to please. Here it tested the client board's own headline figure instead of repeating it, and the deck says why. Forty-five minutes for a work package that is easily five to ten days of work without it. An exhibit from the first deliverable — the sizing funnel, in the consultant's template. The economics behind it, recomputed from the model that sits under the deck. Quality control — three layers, then a loop Nothing reaches the client before three independent passes. Layer 1 — sources. Every claim and every figure traced back to where it came from, the source assessed, each one rated: verified, probable, or not established. Layer 2 — internal audit. An isolated sub-agent reviews the finished document against a purpose-built grid and rebuilds the Excel model from its own inputs. Same family of model, so it catches the obvious: the unsupported claim, the figure that contradicts the model, the page that does not carry its title. Layer 3 — dual external audit. Two models from outside the system, independent of each other and of the one that did the work. No shared bias, and two verdicts to compare. This is where the qualitative control happens. Here the two disagreed — which is itself the finding. Then the loop. The consultant asks the system to integrate the findings and audit again, and ...
adbl_web_anon_alc_button_suppression_t1
まだレビューはありません