AI Judge changed title after evaluation, original title: "AI consultant Hamel Husain calls traditional enterprise AI consulting and deployment practices an organizational anti-pattern"
Hamel Husain draws on more than a decade of hands-on AI work to argue that the familiar pattern of hiring outsiders to build and run enterprise AI systems tends to lock in the very problems it claims to solve, such as missing internal skills and slow decision processes, rather than building lasting capability inside the organization.
Two simultaneous AI narratives right now:
1) You can now do the work of 20 people, learn anything and create anything with AI. Just learn how to use claude.
2) We are investing Billions in forward deployed engineers to help you implement AI b/c its too time consuming and hard for you to do it yourself.
I skew more towards #1, fwiw. But I find these narratives to be incongruent.
Husain notes that companies often turn to third parties because of deeper issues like talent shortages or internal politics, yet those same constraints remain once the engagement ends, leaving teams dependent instead of equipped.
His experience suggests third parties are better used for teaching teams directly and leaning on agents where possible, though no specific client outcomes or timelines back the view in the current discussion.
Many users dismissed claims that AI easily empowers users or replaces consultants as delusional or unrealistic because it still requires engineers and human intervention amid frequent failures.
No Digg Deeper questions have been answered for this story yet.
Indeed: if we actually had AGI we would not need forward deployed engineers.
Two simultaneous AI narratives right now:
1) You can now do the work of 20 people, learn anything and create anything with AI. Just learn how to use claude.
2) We are investing Billions in forward deployed engineers to help you implement AI b/c its too time consuming and hard for you to do it yourself.
I skew more towards #1, fwiw. But I find these narratives to be incongruent.
@HamelHusain explanation for why number two is necessary is obvious if you've ever tried consulting for big companies.
Two simultaneous AI narratives right now:
1) You can now do the work of 20 people, learn anything and create anything with AI. Just learn how to use claude.
2) We are investing Billions in forward deployed engineers to help you implement AI b/c its too time consuming and hard for you to do it yourself.
I skew more towards #1, fwiw. But I find these narratives to be incongruent.
@xlr8harder Oh, I know. I've done this for over a decade. I think that is specifically an anti-pattern for AI.
@HamelHusain explanation for why number two is necessary is obvious if you've ever tried consulting for big companies.
Ive done consulting for a while and I've found that enterprises often hire consultants due to some other problem (talent, org issues, politics)
With AI, this "other problem" is often people aren't using AI in their own workflows, so they lack a mental model and feel like they need outside help.
The next most common problem I see is the same classic talent, org, politics issues where you trust a third party to execute more than your own people.
Either way "implement it for you" feels like an anti pattern because AI should let you get more leverage out of what you have. It should be more "we will teach you" IMO.
I think the whole idea is to make number two less true and number one should be more true over time.
I think an exception to this is upskilling people, to push them into number one so they're more self-sufficient. Having someone implement things for you and walk away seems like an anti-pattern to me, with AI specifically.
@GaryMarcus I genuinely think models are already good enough to replace a lot of consultants right now.
Indeed: if we actually had AGI we would not need forward deployed engineers.
@jdjohnson Low ambition and low agency are not going to be effective with AI, regardless if somebody implements it for you or not.
@HamelHusain I don’t think they’re incongruent, they just refer to different types of work. If you’re solo or own a small business it’s 1). Enterprises are 2).
Been thinking about this a lot, I think both are true at the same time: while an individual is getting more and more leverage it’s still very difficult to build AI systems at enterprise scale.
Individuals at big enterprise companies won’t be able to build proper data infra, permission layers or complex AI systems with evals etc.
@HamelHusain I think both can be true.. to me number 2 is more about how you inject AI/LLMs deep in the enterprise operational workflows that are hard to reach with chat and other tools.
I’ve been so curious about AI and dove into Option 1 upon ChatGPT launching and feel personally empowered and excited and productive. On the other hand, I’m constantly badgered at work by corporate folks who dislike my initiative and primarily want to use outside resources. I think BOTH are true.
@SigsNYC It seems clear to me that the distinction between coding and chat is going away completely, and that these are trending towards super apps. Codex is already like this and many non-technical people are using Claude Code.
@HamelHusain The first is true for curious people with high ambition and high agency. And it doesn’t scale in orgs easily.
The second is true for most employees at most companies.
I think they can both be true.
@HamelHusain it’s been the same with pre-ai tech. You can learn excel and do the work of 20 people with it. Big companies hire consultants to use excel for them.
I agree for coding specifically, as it allows somewhat technical/curious folks to move up the curve and get leverage.
However, I have also been in rooms of ops. folks in the Enteprise who are barely scratching the surface with Chat. Thats why there is currently an obsession with your number 2.l above.
@SigsNYC I think we are already there tbh. I think model capabilities allow people to be way more self-empowered than they might believe (but can't completely replace them either)
It's just a matter of diffusion at this point
@HamelHusain @xlr8harder We building machine gods but can’t beat corporate red tape
@xlr8harder Oh, I know. I've done this for over a decade. I think that is specifically an anti-pattern for AI.
@HamelHusain Good point, ageeed it will take time. We are extremely early in the adoption curve
They are not incongruent.
In scenario 1, the users have both the knowledge and the incentive to build leverage with AI.
In scenario 2, the corporate buyers don't have sufficient knowledge and/or sufficient incentives from their internal users to build leverage with AI themselves (this sounds fancy but its a classic build vs buy vs partner decision and not that deep).
Scenario 1 will slope upward as info spreads (more users with knowledge) and AI native companies gain market share (more users with incentives). We will see more build vs buy/partner, etc...
Scenario 2 is one of the levers AI native cos are are using to spread knowledge (re-write the workflows and tools on AI stacks) and incentives (to expand their marketshare).
@HamelHusain I think it’s just like software - some small teams build the software they need to unlock capacity, some other teams who are better at other things use software built by others, to good effect.