Engineering notes on why an AI-native platform has to be designed that way from the foundation, not added to a stack that already exists.
A consulting engagement crosses seven handovers between first conversation and paid invoice, and most firms run each one in a different tool. Here is what one system has to cover, the questions to ask any vendor, and the mistakes firms make when they consolidate.
Filed under Running a consultancy
A client portal can end the weekly status email, or it can show a client your internal notes. What a consultancy’s clients should see, what they should not, the grey areas to decide per engagement, and the questions to ask before you choose a portal.
The integration tax calculator’s own model, run at three team sizes with its default assumptions. Every line, every total and every input is stated, so each number can be reproduced in the calculator, along with why the share that is not licences falls as a team grows but never drops below half.
Partners ask three questions every week. Are we busy, are we on budget, are we making money? Here is what each number means, a worked example of how a fixed-fee engagement loses margin quietly, and what to track so you see it in week four rather than at month end.
A CRM runs the work before the signature, PSA runs the work after it, and an all-in-one platform tries to hold both on one set of records. What each one does, where each fits by firm size and complexity, what the seams between them cost, and when a dedicated PSA is the right answer.
A sales CRM is built for transactions that end at “closed won”. Consulting runs on engagements that start there. What a consultancy should look for in a CRM, a demo script that tests the handover from sale to delivery, and the questions to put to any vendor.
The AI demo is not the risk. The permission model underneath it is, and it is the thing that decides whether the feature can ever leave the pilot group. Eight questions, product-neutral, with the answer you want to hear for each.
Cyril is a full business platform built by AI, directed by one person. The interesting part is not the speed — it is that the practices which make AI a reliable developer turn out to be the same practices that make software agent-ready.
Licences are the smallest line. Here is an open, assumption-by-assumption model of what it costs to run sales, support, docs, projects and finance in five separate systems — including the line that has only existed since AI arrived.
Every vendor now says their platform is AI-native. Seven questions separate the ones where that is an architectural fact from the ones where it is a feature line — each with a concrete answer you can ask for in a single demo.
Generative AI answers questions; agentic AI takes actions — and actions are where trust is lost. What makes them safe in Cyril isn’t a guardrail prompt: it’s permission checks on the server, approval before anything runs, and a record of every run.
Cyril gives its core records an ai_context serializer — a deterministic, schema-aware view built for the model, not the database. It’s the unglamorous core of what makes AI grounding actually work.
Bolted-on AI looks impressive in demos and falls apart in workflows. The reason is structural. Here’s the engineering decision that separates AI-native from AI-added-on.
Four ways in. A post can sit under more than one, because the integration tax is as much an AI argument as an operations one.
Why an AI-native platform has to be designed that way from the foundation: one data model, one context surface, one set of guarantees.
Open the topicWhat has to be true underneath before software can safely take actions on your behalf: permissions, an audit trail, and a way back.
Open the topicHow to evaluate a business platform, and what a stack of disconnected tools actually costs. Written for the person who has to decide.
Open the topicNotes from building Cyril: the decisions, the failures, and what they taught us about software an agent can operate.
Open the topicFor consulting firms: winning engagements, delivering them on budget and getting paid, and what the software underneath has to do at each step.
Open the topicOne platform for sales, projects, support, docs and financials, with the AI underneath rather than bolted on. Join the waitlist to be among the first to use it.
