posted 1st September 2026
Agentic AI is collapsing the time it takes your organisation to act on a decision. The work in the middle, building, integrating, rolling out, is getting dramatically faster, and the verdict on a decision now lands sooner, and more publicly, than it used to. That lag is shrinking. This sounds like good news, right? And partly it is. But that lag between deciding something and living with the consequences has been quietly protecting leadership teams for decades. Now, it's going away.
For many large organisations, whether a bank, a manufacturer, a hospital trust or a retailer, a leadership team makes a handful of genuinely consequential decisions, five to ten that really matter, playing out over years. What to build. What to fund. Which projects to bring to life, and how to structure the business around them. Historically a poor call could take years to show itself, and that gave you room to adjust, to reorganise, to move on before the bill arrived. Slow execution was a cushion, and weak decisions aged quietly.
Two things change that. Any decision that depends on shipping something now plays out in months, not years. And the ones that genuinely do take years, entering a new market, say, still show their hand sooner, and more publicly, because slow delivery no longer buries a weak call. It just delays the reveal.
So, the cushion goes. A decision that wasn't properly thought through now gets exposed while you're still in the job, in front of customers and the market. A good call pays off faster. A bad one costs you faster, and in public.
I've seen work that would once have taken a developer a week, call it forty hours, come back from an AI agent in an evening. Call it four. That's one person at roughly ten times the throughput.
It doesn't stop at one desk. A team of eight clears a fortnight's build in an afternoon. A hundred-person department turns a year on the roadmap into a quarter. A thousand-person organisation turns a three-year programme into a one-year one. The gap between a decision and its consequences is closing at every level at once, and it's not going to open back up.
Zillow is the cleanest warning. Zillow Offers launched in 2018 and ran for about three years as a cautious operation. What sank it was older thinking: a 2019 strategy to turn home-buying into a very large business, thousands of purchases a month, on the assumption its pricing models could value houses well enough to buy, renovate and resell at scale. In 2021 it took the brakes off, buying more homes in a single quarter than in the previous eighteen months combined, into a cooling market, its own confidence in the model already slipping. Nothing slowed the buying as the forecasts drifted. That November it shut the division down, wrote off more than $500m and cut around 2,000 jobs, a quarter of the company. Opendoor and Offerpad ran the same model through the same market and survived, so this was a governance failure, not a verdict on algorithmic pricing. The algorithm did exactly what it was asked. It executed a 2019 bet at a pace no human team could match, so when the assumption broke the loss landed in one quarter instead of bleeding out over years. The build worked. The thinking it ran on did not.
DPD is the same shape at a different speed. The parcel firm had run a chat assistant for years without a public incident. A system update on 18 January 2024 let the language model behind it go off-script. Egged on by a frustrated customer, it swore, called itself useless, branded DPD "the worst delivery firm in the world", and knocked out a poem saying the same. The customer posted the screenshots to X, and within a day or two the post had more than a million views and DPD had pulled the AI. Where Zillow took three years, DPD took two days. And it took no grand strategy to get there, just a release nobody had checked.
A redeployment, not a restructure
The instinct may be to reach for structure: a new function, a hire, a steering committee. It's likely you don't need any of that. What you do need to change is where some people sit and when they're brought in. Most large organisations already employ the ones who can help, the agile coaches, the transformation people, currently pulled in downstream and late to tune delivery teams. The value now (and in most cases always has been) is upstream and early, while the big decisions are still being shaped. This is likely a redeployment of people you already pay. If you've got nobody like that, what you're looking for is people who can build an options set, design a cheap test, and run an honest decision meeting, not a particular job title. Be clear about the limit, though: they can't do the thinking for you, any more than they did it for the delivery teams. What they can do is bring the right methods, challenge a decision before it's locked in, and help you build a way of working that actually gets you the outcomes you're after.
Do the thinking before you spend the money
There's no trick to this. You do more of the work before you commit, not after. Name the assumption that sinks the whole thing if it's wrong. Run the cheapest test you can against it before serious money goes in. Ask for a real set of options, not one recommendation with two strawmen propped up next to it. And put in a genuine decision point, one where "not this one" is an allowed answer and something actually gets stopped.
One more thing:describe the change you're after, not the thing to be built. "It lets me do X", not "it must do X". David J. Anderson calls this Fit for Purpose. Worth noting he's also the man behind the Kanban Method, which organisations everywhere adopted to speed delivery up, and he told us to lead with purpose too. That's the half most people skipped.
Ask yourself: how many ideas does your leadership team kill on purpose, before committing, in a normal year? If the answer is close to none, you don't have a decision process, you have an approval process. A serious pipeline discards a good deal of what enters it. If nothing ever gets stopped, either every idea reaching the top is excellent, or nobody is testing them.
Who this catches out
Smaller, faster businesses already work this way. They test an idea in front of real customers before betting on it, because they can't afford not to. For some large organisations, execution stays genuinely hard because regulators, legacy systems and capital cycles slow it down. That's real, not a failure of will. But the same problem still applies. Speed delivery up inside a fixed requirements-and-business-case model and a weak call doesn't get any better, it just hits the market faster, and with more confidence, because look how quickly it shipped.
One thing to try this week
For a decision you can actually test: take the biggest one the team feels sure about. Before the next chunk of investment goes in, write down the single assumption it rests on, the one that kills it if it's wrong. Agree now what evidence would make you stop, then send someone who can run a cheap test to go and get it inside a fortnight. For the calls you can't test out loud, an acquisition, say, you do it differently: a proper pre-mortem, a red team paid to argue the other side, kill criteria written down in advance, the money released in stages.
If the team can't agree what would change its mind, that's the finding. You're committed to something you haven't tested.
The next weak decision your organisation makes will be visible faster than the last one, and more publicly. Execution is becoming a commodity; the advantage is in choosing well, and most of the people who can help you do that are already on your payroll.
So, from here, two kinds of organisation pull apart. One treats AI's arrival as a reason to change how it works: slowing right down on discovery and ideation, redesigning the processes and workflows that feed delivery, putting real rigour in front of a decision before it's committed. The other hands everyone an LLM licence, writes a policy document, and changes nothing else, so it makes the same untested decisions it always did, only faster and in public. The tools are the same for both. The question is whether you're willing to redesign the way the work gets done. Which one will yours be?