AI consultant who ships: from strategy to production

An AI consultant works out where AI will save you time or money, picks the first thing to build, and gets it running safely. I’m Mo Tawfik Khalifa, chief product officer and forward deployed engineer at Instruxi, and I do the consulting and then build what we decide, all the way to production.

The build is where most AI projects fail. In August 2025, MIT’s NANDA initiative reported that only about 5% of the AI pilots it studied reached rapid revenue growth, and that tools bought from specialist vendors worked about 67% of the time, against about a third as often for tools companies built themselves (Fortune on the report, The GenAI Divide). Choosing well and shipping carefully matter more than the model.

Worried it ends as a demo nobody owns?

Three worries operations leaders raise about AI projects, in their own words where they said it in public, and what I do about each.

• “We’ve seen dozens of demos this year. Maybe one or two are genuinely useful. The rest are wrappers or science projects.”

  A CIO in MIT NANDA’s report, The GenAI Divide, 2025 (PDF)

  I start with one workflow that already costs your team hours, and we write down its number before I build: hours a week, tickets a day or days to close. It goes live in production on your real data, and we measure the same number again. You get a before and an after, not a demo.

• “projects that are declared a success by management before they are actually done, and no one is officially responsible for maintaining it.”

  An engineer on Hacker News, February 2026

  Before I build, we name the person on your side who owns it after launch. They see the work on a live link every week, and they get a written handover: how it works, what the AI may do alone, and what to check each week. You are not dependent on me.

• Security or compliance stops it at the last step.

  The security review starts in week one, not after the demo. We agree what data the AI may read, where that data is sent, what it may do alone and what needs a person, and every action it takes is logged. It starts read-only and gets one action at a time.

What you get:

• One workflow live in production, with its before and after number.
• A named owner on your side, and a written handover so you are not dependent on me.
• The code, the repo and the accounts in your name from day one.
• The price fixed in writing after the first call, before work starts.
• A live link you can click every week, and you talk to me, the person who builds it.

What does an AI consultant do?

A good AI consultant does five things, in this order:

1. Finds the task. One job that repeats, costs real hours, and has a number you can measure.
2. Checks the data. Where it lives, who owns it, and what the AI is allowed to read.
3. Decides buy or build. Buy the commodity part, build the part that is yours.
4. Sets the limits. What the AI may do alone, what needs a person, and how you prove afterwards what it did.
5. Gets it to production, and hands it to someone on your team who owns it.

Most stop after step three. The deliverable is a report or a slide deck, and someone else is left to wire it into real systems, add identity and permissions, and keep it from doing something expensive. That gap is the job I do. I wrote up how a forward deployed engineer differs from a consultant: the consultant hands over a plan, the forward deployed engineer stays until it runs.

What do I do instead of slides?

I come from coding and design, and I ship with AI at every step. Every engagement runs through three moves.

• Assess. I sit with your team and your customers, read the tickets and the data, and write the brief back on one page: the one number, what we build, what we buy and what we cut.
• Prototype in days. Something your team can press and break, on your real data. When I built Laqtaa alone with AI, six milestones were built and deployed in 53 minutes on the first day. The weeks after that were the real work.
• Ship with guardrails. Identity, least privilege, a check before every action, spend and loop limits, and a record you can prove. I lead product for GVNR at Instruxi, a free, source-available guard that answers allow, deny or ask a person before an agent acts. It works with any AI agent, including Claude Code, Cursor, Codex, Gemini CLI and any MCP client. The framework is in AI agent governance.

Before Instruxi I led a 24-person team of developers, designers and producers at Napster, and led technology at Consortium9 as Director, then VP, then CTO. Twenty years of live systems taught me to load the fallback before going on air.

Pick your situation on the board below to see which engagement fits and what the first 30 days look like. With the defaults: a support agent on your help articles, where mistakes cost money, needed this quarter, is a build sprint that starts read-only.

AI consulting services and AI implementation services: what does an engagement look like?

Three shapes. The board above suggests one, and we settle it on a first call.

• A short assessment. I sit with your team, read the data and the tickets, and prototype the most promising task to test it. You get a one-page call: build, buy or stop, with the first build scoped. It ends with something you can press, not a deck.
• A build sprint. One use case, from prototype on your real data to production: read-only first, then actions one at a time behind a check, a person watching at launch, and a test for every claim the product makes.
• Fractional leadership. When AI is part of your product, someone has to own the product calls every week. I lead product for one to four days a week and build alongside your team. That is the fractional CPO work.

Every shape ends the same way: docs, playbooks and a named person on your side who owns it after me.

What have I built that looks like your problem?

I built all of these, but most of my work at Instruxi is under NDA, so I can’t name the clients. These are the builds I have written up, with the proof you can open.

If you are connecting AI agents to your own tools, I also wrote up how to build and ship an MCP server, and what an AI agent is for the people on your team who need the plain version.

How much does an AI consultant cost?

AI consultants charge by the day, by a monthly retainer for set days a week, or a fixed price for a defined piece such as an assessment or a prototype. The scope, the days a week and how much gets built set the price.

I price each engagement after a first call, from the shape, the length and how much I build.

How do you choose an AI consultant? Seven red flags

1. No working demo in the first weeks. If all you have after a month is slides, you are paying for the easy half.
2. Building before buying. In the MIT report, tools bought from specialist vendors worked far more often than ones built in house. Ask what they would buy, and why the rest has to be custom.
3. No answer to “what happens when it is wrong?” They should name what the AI may do alone, what needs a person, and how you will prove afterwards what it did.
4. Write access to everything on day one. Start read-only and add one action at a time. My governance framework has the order.
5. A promised accuracy or resolution rate before they have seen your data. The number depends on your articles, your data and your permissions.
6. A tool they resell, not said up front. Ask what they earn from each vendor they recommend.
7. Nobody owns it after they leave. Ask who on your team runs it in month four, and what they will be handed.

And one test: ask them to show you something they built and shipped, that you can open. Mine are linked above.

Questions people ask

What does an AI consultant do?

Finds where AI can save time or make money in your business, picks the first use case, decides what to buy and what to build, and plans the data, security and people around it. The good ones stay until it works in production.

Do I need an AI consultant or an AI developer?

If you know the task and have the data, you need someone to build it. If you do not know which task is worth it, you need the consulting first. I do both, so the assessment turns into the build without a handover.

What is an AI readiness assessment?

A check of whether one task in your business is ready for AI: is it worth the hours, is the data reachable and owned, and what may the AI do alone. Mine is the short assessment above: I read the data and the tickets, prototype the most promising task on your real data, and give you a one-page call to build, buy or stop. Good generative AI consulting ends with something your team can press, not a readiness score.

What is the difference between an AI consultant and a forward deployed engineer?

A consultant studies the problem from outside and hands over a plan. A forward deployed engineer ships the working software inside your systems and stays until it runs. I work as both.

How long until something is in production?

A prototype on your real data takes days. The slow part is hardening: identity, permissions, limits and an audit trail. That is where most pilots stall, and it is the part I plan for from week one.

Can the AI take actions, or only answer questions?

Both, in that order. I start agents read-only, then add one action at a time behind a check that runs outside the model: allow, deny or ask a person, with a record of every decision.

Do you work with UK and US companies, and remotely?

Yes. I work with UK teams and teams elsewhere, remote or on site. I have led teams across countries, on site and remote, for more than eleven years.

1. 30-minute call. Your goal, your users, what is stuck.
2. A written plan in 48 hours. What to build first, how long it takes, the risks.
3. Start, or not. Use the plan with your team or with me. I reply to every email within one working day.

If you need AI scoped and built, write to m@tawfik.tv with what you want it to do and where your data lives.