Over the past few weeks, two thought­ful arti­cles cut through the relent­less AI hype and gave me pause for reflec­tion. In The Econ­o­mist, Ethan Mol­lick warned that “the IT depart­ment is where AI goes to die.” His point is sharp: AI is a pro­found­ly strange, risky, and pow­er­ful tech­nol­o­gy — a next-word pre­dic­tor that some­how writes code, offers strate­gic coun­sel, or even sim­u­lates empa­thy. Yet many orga­ni­za­tions are smoth­er­ing its poten­tial by forc­ing it into the rigid mold of tra­di­tion­al enter­prise software.

Around the same time, the Finan­cial Times pub­lished a piece not­ing that while investors are bet­ting on AI-fueled chaos and dis­rup­tion, his­to­ry tells a dif­fer­ent story. Past tech­no­log­i­cal revolutions—from the PC to the inter­net and cloud—rarely wiped out incum­bents. Savvy estab­lished play­ers adapt­ed, inte­grat­ed the new capa­bil­i­ties, and often emerged stronger.

Read­ing these togeth­er crys­tal­lized some­thing I’ve been observ­ing in our work at Datarel­la and across the broad­er tech land­scape: the most prob­a­ble path for AI in busi­ness is nei­ther a dystopi­an job apoc­a­lypse nor a chaot­ic upend­ing of entire indus­tries. It’s a prag­mat­ic, evo­lu­tion­ary inte­gra­tion — one that rewards orga­ni­za­tions will­ing to embrace AI’s inher­ent “weird­ness” while build­ing solid foun­da­tions to pre­vent disorder.

As some­one who has spent decades build­ing com­pa­nies and help­ing enter­pris­es nav­i­gate dig­i­tal trans­for­ma­tion, I believe this bal­anced view is cru­cial. AI won’t replace every­thing overnight, but it will reshape how we work — if we let it.

Why AI So Often “Dies” in Traditional IT Settings

Mollick’s diag­no­sis rings espe­cial­ly true because we see this pat­tern repeat­ed­ly in enter­prise envi­ron­ments. AI isn’t deter­min­is­tic soft­ware with pre­dictable, repeat­able out­puts. It’s gen­er­a­tive, high­ly context-dependent, and fre­quent­ly sur­pris­ing. When hand­ed over to IT teams whose pri­ma­ry man­dates are secu­ri­ty, com­pli­ance, uptime, and cost con­trol, the instinc­tive reac­tion is under­stand­able but often counterproductive:

• Wrap­ping every exper­i­ment in lengthy approval processes
• Demand­ing detailed ROI pro­jec­tions before any mean­ing­ful pilot
• Forc­ing AI into lega­cy tech stacks with­out rethink­ing under­ly­ing workflows
• Pri­or­i­tiz­ing only the safest, most obvi­ous use cases

The out­come? Count­less pilots that never scale. Recent analy­ses, includ­ing Deloitte’s 2026 State of AI in the Enter­prise, show that while employ­ee access to AI tools has explod­ed, the move from exper­i­men­ta­tion to full pro­duc­tion remains lim­it­ed. Issues like poor data qual­i­ty, skills gaps, and over­ly cau­tious gov­er­nance con­tin­ue to cre­ate friction.

Har­vard Busi­ness Review has noted a sim­i­lar phe­nom­e­non: wide­spread AI usage paired with dis­ap­point­ing returns, with adop­tion often stalling at the inte­gra­tion stage. The core mis­take isn’t poor execution—it’s treat­ing AI like just anoth­er CRM or ERP mod­ule rather than a fun­da­men­tal­ly new way of think­ing and working.

History Offers Reason for Optimism

The Finan­cial Times arti­cle pro­vides a reas­sur­ing coun­ter­point. Tech­nol­o­gy rev­o­lu­tions rarely play out as pure cre­ative destruc­tion. Incum­bents who invest in com­ple­men­tary capabilities—new skills, redesigned process­es, and updat­ed orga­ni­za­tion­al structures—tend to adapt and thrive.

In 2026, I expect the real win­ners won’t be only the flashy AI-native star­tups. They will be estab­lished com­pa­nies that intel­li­gent­ly com­bine their deep domain exper­tise and pro­pri­etary data with AI’s capa­bil­i­ties. Those who redesign work­flows for gen­uine human-AI col­lab­o­ra­tion (some­times called “co-intelligence”) and scale thought­ful­ly from pilots to enterprise-grade agen­tic sys­tems will gain the edge.

Reports from PwC and oth­ers speak of a “dis­ci­plined march to value”: clear strate­gies, mea­sur­able out­comes, and gov­er­nance frame­works that pro­tect with­out suf­fo­cat­ing innovation.

What the Most Likely Future of AI in Business Looks Like

Look­ing ahead to late 2026 and 2027, here’s the tra­jec­to­ry I con­sid­er most probable:

1. Scal­ing from pilots to pro­duc­tion — More orga­ni­za­tions will move a sig­nif­i­cant­ly high­er share of AI projects into live use, par­tic­u­lar­ly through agen­tic AI sys­tems that han­dle multi-step work­flows autonomously.
2. The J‑curve of pro­duc­tiv­i­ty — Expect ini­tial peri­ods of flat or even neg­a­tive returns as com­pa­nies rewire process­es and roles. Once the com­ple­men­tary changes (new data pipelines, deci­sion pro­to­cols, and team struc­tures) are in place, gains should accel­er­ate sharply.
3. Gov­er­nance matur­ing — Robust frame­works for respon­si­ble agen­tic AI, data qual­i­ty, and risk man­age­ment will become stan­dard. “Shad­ow AI” will grad­u­al­ly decline as secure, enterprise-ready plat­forms improve.
4. Incum­bents lever­ag­ing their data moats — Orga­ni­za­tions with clean, well-governed data and strong domain knowl­edge — espe­cial­ly in reg­u­lat­ed or com­plex industries—will often out­per­form pure AI disruptors.

This isn’t a utopi­an rev­o­lu­tion or a total fail­ure. It’s an evo­lu­tion­ary trans­for­ma­tion, pro­vid­ed we avoid the trap of over-standardizing AI too early.

Five Practical Principles for Embracing AI’s Weirdness

Draw­ing from Mollick’s insights, his­tor­i­cal pat­terns, and the lat­est enter­prise reports, here are the prin­ci­ples I believe forward-thinking lead­ers should adopt:

1. Delib­er­ate­ly embrace the weird­ness — Cre­ate space for teams to exper­i­ment and dis­cov­er unex­pect­ed appli­ca­tions. Encour­age “labs” or crowd­sourced explo­ration. Treat AI as a cre­ative col­lab­o­ra­tor rather than a sim­ple automa­tion engine.
2. Invest in rock-solid data foun­da­tions — Data qual­i­ty and gov­er­nance remain the biggest bar­ri­ers. With­out trust­wor­thy, well-integrated data, even the most advanced mod­els pro­duce unre­li­able results. This is an area where spe­cial­ized exper­tise in uni­fy­ing silos and build­ing real-time, com­pli­ant pipelines makes a deci­sive difference.
3. Redesign work­flows for human-AI co-intelligence — The goal isn’t to auto­mate jobs out of exis­tence but to aug­ment human strengths. Let peo­ple focus on judg­ment, cre­ativ­i­ty, and rela­tion­ships while AI han­dles analy­sis, draft­ing, and rou­tine tasks.
4. Deploy gov­erned, secure agen­tic sys­tems — Autonomous agents rep­re­sent the next fron­tier, but they require thought­ful orches­tra­tion, threat mod­el­ing, and com­pli­ance built in from the start.
5. Mea­sure what truly mat­ters and iter­ate patient­ly — Look beyond van­i­ty met­rics. Track real busi­ness impact—revenue, cost effi­cien­cy, cus­tomer outcomes—and accept that returns often fol­low a J‑curve.

Reflections from the Trenches

At Datarel­la, we’ve been help­ing orga­ni­za­tions move past the hype and pilot pur­ga­to­ry for years. Our focus on secure AI agent devel­op­ment, full-stack mod­ern­iza­tion, privacy-preserving archi­tec­tures, and (where appro­pri­ate) decen­tral­ized approach­es is designed pre­cise­ly for this moment: enabling com­pa­nies to har­ness AI’s strange power with­out invit­ing chaos.

Whether it’s build­ing production-ready autonomous agents, cre­at­ing reli­able data plat­forms, or inte­grat­ing AI into com­plex lega­cy envi­ron­ments, the key is com­bin­ing tech­ni­cal depth with prac­ti­cal busi­ness judgment.

The future of AI in busi­ness isn’t about tear­ing down your exist­ing struc­tures or gam­bling on total dis­rup­tion. It’s about evolv­ing how your orga­ni­za­tion learns, decides, and cre­ates value—by thought­ful­ly embrac­ing AI as the odd, pow­er­ful tool it is, while strength­en­ing the data, gov­er­nance, and cul­tur­al foun­da­tions it requires.

If you’re ready to move from inter­est­ing pilots to scal­able impact—without let­ting AI “die in IT”—I’d be happy to explore how we can sup­port your journey.

Let’s con­nect.

Agentic AI AI Datarella Ethan Mollick Generative AI RAAY RE